1 00:00:02,720 --> 00:00:14,000 Speaker 1: Bloomberg Audio Studios, Podcasts, radio News. 2 00:00:18,520 --> 00:00:21,680 Speaker 2: Hello and welcome to another episode of the Odd Lots Podcast. 3 00:00:21,760 --> 00:00:23,880 Speaker 3: I'm Joe Wisenthal and I'm Tracy Alloway. 4 00:00:24,200 --> 00:00:25,040 Speaker 4: Racy, I have a question. 5 00:00:25,280 --> 00:00:27,640 Speaker 2: I know you spent a lot of your youth overseas 6 00:00:27,800 --> 00:00:30,920 Speaker 2: my youth. Did you ever go to many baseball games 7 00:00:30,960 --> 00:00:31,400 Speaker 2: as a kid. 8 00:00:31,920 --> 00:00:34,440 Speaker 3: Yeah, so I was in Chicago for a few years, 9 00:00:34,479 --> 00:00:36,400 Speaker 3: so I went to the Cups games, and then I 10 00:00:36,479 --> 00:00:40,480 Speaker 3: was in Japan, and the baseball scene in Japan is amazing, 11 00:00:40,600 --> 00:00:43,800 Speaker 3: like the best vibes of a live sports event that 12 00:00:43,960 --> 00:00:46,320 Speaker 3: I've ever ever witnessed. Her encounter. 13 00:00:46,440 --> 00:00:48,159 Speaker 2: I was just talking to someone about this last night. 14 00:00:48,159 --> 00:00:51,080 Speaker 2: I've always wanted to go to a Japanese baseball record. 15 00:00:51,159 --> 00:00:54,800 Speaker 2: That's like, if we ever do a live show in Tokyo, 16 00:00:55,440 --> 00:00:57,840 Speaker 2: let's try to schedule it during baseball season, because it 17 00:00:58,040 --> 00:01:00,880 Speaker 2: is like I sort of think that's a bucket list thing. 18 00:01:01,160 --> 00:01:03,600 Speaker 2: But anyway, the reason I asked this question is I 19 00:01:03,640 --> 00:01:06,120 Speaker 2: have this really vague memory as a child going to 20 00:01:06,160 --> 00:01:09,200 Speaker 2: see Detroit Tigers games with my grandfather, like probably when 21 00:01:09,200 --> 00:01:11,400 Speaker 2: I was like maybe these memories are probably from when 22 00:01:11,400 --> 00:01:14,000 Speaker 2: I was younger than like six or five, but there 23 00:01:14,080 --> 00:01:17,160 Speaker 2: used to be a non trivial number of people who 24 00:01:17,160 --> 00:01:20,000 Speaker 2: would go to the games and they would keep score, 25 00:01:20,080 --> 00:01:23,600 Speaker 2: and they would write down every single a bat and 26 00:01:23,640 --> 00:01:25,240 Speaker 2: they the outcome of every single. 27 00:01:25,160 --> 00:01:27,560 Speaker 3: Your I've seen that, not in person, but like maybe 28 00:01:27,720 --> 00:01:28,920 Speaker 3: movies or something like that. 29 00:01:29,000 --> 00:01:31,240 Speaker 2: It doesn't really happen anymore, like I never see it. 30 00:01:31,280 --> 00:01:33,440 Speaker 2: There are may be a few, like old timers who 31 00:01:33,480 --> 00:01:36,080 Speaker 2: still has a hobby or habit do that, but it 32 00:01:36,160 --> 00:01:39,240 Speaker 2: was like a non insignificant number of people. And it's 33 00:01:39,280 --> 00:01:40,959 Speaker 2: interesting to me. You know, I've been to a couple 34 00:01:41,000 --> 00:01:44,880 Speaker 2: of soccer games this year. There's no equivalent way you 35 00:01:44,920 --> 00:01:47,960 Speaker 2: could do that, right, because it's like baseball is filled 36 00:01:47,960 --> 00:01:50,720 Speaker 2: with all of these discrete events. The picture, who is 37 00:01:50,720 --> 00:01:51,920 Speaker 2: the picture? Who is the batter? 38 00:01:52,680 --> 00:01:52,880 Speaker 5: Hit? 39 00:01:53,120 --> 00:01:56,240 Speaker 2: Not a hit, single, not a strikeout, walk, et cetera. 40 00:01:56,400 --> 00:01:58,559 Speaker 2: Like what would even be the equivalent to soccer? 41 00:01:58,720 --> 00:02:01,360 Speaker 3: So this has been a long running debate in soccer. 42 00:02:01,400 --> 00:02:03,960 Speaker 3: And I remember when moneyball came out, and you know, 43 00:02:04,040 --> 00:02:07,160 Speaker 3: sports analytics became a big thing, especially for baseball because 44 00:02:07,200 --> 00:02:09,680 Speaker 3: as you point out, it's these sort of discrete events 45 00:02:09,720 --> 00:02:13,080 Speaker 3: that have a lot of statistics embedded in them. A 46 00:02:13,120 --> 00:02:15,519 Speaker 3: lot of people were saying that soccer. You could never 47 00:02:15,680 --> 00:02:18,040 Speaker 3: use data analytics in the same way for soccer, Like 48 00:02:18,080 --> 00:02:20,919 Speaker 3: it's too chaotic, it's too fluid, there's too many variables, 49 00:02:21,000 --> 00:02:24,440 Speaker 3: there's not enough goals. That complaint comes up a lot 50 00:02:24,480 --> 00:02:26,880 Speaker 3: whenever we talk about are we going to say soccer 51 00:02:26,960 --> 00:02:28,480 Speaker 3: or football? By the way, you know what, let's just 52 00:02:28,480 --> 00:02:31,280 Speaker 3: say soccer. Okay, all right, I'll try. 53 00:02:31,440 --> 00:02:33,800 Speaker 2: We can say football. I actually don't feel strongly about 54 00:02:33,800 --> 00:02:34,079 Speaker 2: this one. 55 00:02:34,200 --> 00:02:39,160 Speaker 3: But that said, we do see soccer analytics on the rise, right, 56 00:02:39,240 --> 00:02:41,800 Speaker 3: Like now we get all these stories about like tiny 57 00:02:41,919 --> 00:02:46,160 Speaker 3: clubs that are using data to source you know, specific 58 00:02:46,160 --> 00:02:51,200 Speaker 3: players in very moneyball style. Always we obviously have prediction markets. Yeah, 59 00:02:51,280 --> 00:02:53,400 Speaker 3: people are doing a lot of sports betting, and so 60 00:02:53,520 --> 00:02:56,840 Speaker 3: all the analytics seem to be like becoming more important. 61 00:02:56,840 --> 00:02:59,040 Speaker 3: And I will just say, I've read this crazy stat 62 00:02:59,400 --> 00:03:02,400 Speaker 3: right before we came on from the law firm Morgan Lewis, 63 00:03:02,560 --> 00:03:05,040 Speaker 3: and they were saying, in the twenty twenty six FIFA 64 00:03:05,240 --> 00:03:10,480 Speaker 3: World Cup match based data is going to be something 65 00:03:10,680 --> 00:03:13,839 Speaker 3: like so it's one hundred and four matches generating more 66 00:03:13,880 --> 00:03:18,440 Speaker 3: than ninety petabytes of data, yeah, which is a forty 67 00:03:18,480 --> 00:03:21,839 Speaker 3: five fold increase over the volume produced in the last 68 00:03:21,880 --> 00:03:24,640 Speaker 3: World Cup in twenty twenty two, you know, stunning. 69 00:03:24,720 --> 00:03:26,840 Speaker 2: So this stories is a question, and I know we're 70 00:03:26,840 --> 00:03:28,720 Speaker 2: going to get to this in the conversation, and it 71 00:03:28,800 --> 00:03:33,000 Speaker 2: almost is like a philosophical question, which is, okay, we 72 00:03:33,040 --> 00:03:35,360 Speaker 2: see the game of soccer is like very fluid, right, 73 00:03:35,400 --> 00:03:37,640 Speaker 2: there's no we just said there's a fewer discrete events, 74 00:03:37,640 --> 00:03:41,720 Speaker 2: like in theory, is something that's fluid a series of 75 00:03:41,760 --> 00:03:46,240 Speaker 2: like microscopic discrete events. You know, could you get a 76 00:03:46,360 --> 00:03:50,160 Speaker 2: million frames per second and actually turn it into discrete events? 77 00:03:50,200 --> 00:03:53,000 Speaker 2: Like that is sort of like an interesting question to mean. 78 00:03:53,320 --> 00:03:56,520 Speaker 2: A parallel that I think of in this conversation is 79 00:03:57,520 --> 00:04:00,280 Speaker 2: chess is like all discrete events, and that it's seen 80 00:04:00,320 --> 00:04:02,520 Speaker 2: it's like a very difficult thing to crack. But then 81 00:04:03,080 --> 00:04:04,440 Speaker 2: like over twenty five years ago. 82 00:04:04,400 --> 00:04:05,920 Speaker 3: Now it's like, yeah, of course. 83 00:04:05,800 --> 00:04:07,800 Speaker 2: They cracked it. But then you would think, okay, well, 84 00:04:07,800 --> 00:04:10,280 Speaker 2: like what is the opposite of chess, which would be language, 85 00:04:10,320 --> 00:04:12,600 Speaker 2: And you say, okay, you can't. That's fluid, that's all 86 00:04:12,640 --> 00:04:15,880 Speaker 2: over the place, And yet computers seemed to be understanding 87 00:04:15,960 --> 00:04:16,839 Speaker 2: language pretty good. 88 00:04:17,000 --> 00:04:21,120 Speaker 3: My inner luod Iite says, there must still be a secret, 89 00:04:21,240 --> 00:04:24,800 Speaker 3: like unmodel leable you can't talk when it comes to 90 00:04:24,839 --> 00:04:28,120 Speaker 3: like football, the beautiful game. But you're right that technology 91 00:04:28,200 --> 00:04:29,600 Speaker 3: may prove me wrong very quickly. 92 00:04:29,800 --> 00:04:33,880 Speaker 2: Well, as you mentioned, you know, soccer analytics growing, and 93 00:04:33,960 --> 00:04:37,240 Speaker 2: obviously the interest now is for obvious reasons. There's all 94 00:04:37,279 --> 00:04:40,000 Speaker 2: the betting on the World Cup. People talk about the 95 00:04:40,200 --> 00:04:43,080 Speaker 2: x G of like I don't know a situation or 96 00:04:43,120 --> 00:04:44,680 Speaker 2: a player the expected goals. 97 00:04:44,880 --> 00:04:47,400 Speaker 3: Now we have body posing analytics as well, which. 98 00:04:47,320 --> 00:04:51,720 Speaker 2: Yeah, and when our producer Kale introduced me, I hadn't 99 00:04:51,760 --> 00:04:55,640 Speaker 2: seen them before earlier this year, like the momentum charts 100 00:04:55,760 --> 00:04:59,680 Speaker 2: that show like you know how dominant a team is 101 00:04:59,680 --> 00:05:02,120 Speaker 2: in given moment, you see it going waves and stuff 102 00:05:02,120 --> 00:05:04,200 Speaker 2: like that. So there clearly is a lot of work 103 00:05:04,240 --> 00:05:06,479 Speaker 2: being done. I don't know how much it works. I 104 00:05:06,480 --> 00:05:10,360 Speaker 2: don't know how much like momentum charts consistently predict who 105 00:05:10,440 --> 00:05:12,840 Speaker 2: is going to be the winner. But just this question 106 00:05:12,960 --> 00:05:17,839 Speaker 2: of like the ability to model deeply fluid things the 107 00:05:17,880 --> 00:05:20,880 Speaker 2: beautiful game, like it's an art, right, Like can computers 108 00:05:20,920 --> 00:05:23,800 Speaker 2: actually model this? And then like if they can, what 109 00:05:23,839 --> 00:05:25,680 Speaker 2: does that say about the ability to model a bunch 110 00:05:25,680 --> 00:05:28,279 Speaker 2: of other things? Strikes me as just like a very 111 00:05:29,320 --> 00:05:31,320 Speaker 2: relevant question right now, because it's the World Cup, but 112 00:05:31,360 --> 00:05:34,839 Speaker 2: also relevant question in general about market for markets and 113 00:05:34,880 --> 00:05:39,320 Speaker 2: finance exactly. So I'm curious to know like how it works, 114 00:05:39,920 --> 00:05:43,279 Speaker 2: about how the moneyball revolution, where we are in the 115 00:05:43,400 --> 00:05:46,080 Speaker 2: arc with soccer anyway, I'm really excited to say we 116 00:05:46,160 --> 00:05:49,200 Speaker 2: really do have two perfect guests because they do sit 117 00:05:49,360 --> 00:05:52,000 Speaker 2: right in this space and also at the intersection of 118 00:05:52,040 --> 00:05:54,320 Speaker 2: everything that we're talking about. We're gonna be speaking with 119 00:05:54,560 --> 00:05:58,400 Speaker 2: yours Becker's. He is a soccer analytics consultant, the professional 120 00:05:58,440 --> 00:06:02,120 Speaker 2: soccer Analytics consultant for the decade. As well as Mike Tracy, 121 00:06:02,240 --> 00:06:05,240 Speaker 2: he's the head of risk at Apex Fintech Solutions. He 122 00:06:05,320 --> 00:06:08,560 Speaker 2: used to be a volatility arbitrage trader at Peak six 123 00:06:08,760 --> 00:06:11,920 Speaker 2: and he also does soccer analytics for the club FC. 124 00:06:12,120 --> 00:06:15,240 Speaker 2: So literally the two perfect guests to talk about some 125 00:06:15,400 --> 00:06:18,440 Speaker 2: of these questions that are arising right now. So Mike 126 00:06:18,520 --> 00:06:20,560 Speaker 2: and yours, thank you both so much for coming on 127 00:06:20,600 --> 00:06:21,200 Speaker 2: the podcast. 128 00:06:21,560 --> 00:06:24,160 Speaker 6: Thank you for having us is a great introduction. Yeah, 129 00:06:24,200 --> 00:06:24,679 Speaker 6: thank you. 130 00:06:24,680 --> 00:06:27,400 Speaker 2: You know, Mike, let me start with you question. A 131 00:06:27,480 --> 00:06:32,040 Speaker 2: lot of people I think, have, for good reason an 132 00:06:32,120 --> 00:06:37,200 Speaker 2: intuitive feel that like sports analytics and trading markets are 133 00:06:37,240 --> 00:06:40,720 Speaker 2: adjacent ideas that like, we know this, we know that 134 00:06:40,760 --> 00:06:44,200 Speaker 2: the a lot of the prop shops and market makers 135 00:06:44,279 --> 00:06:47,720 Speaker 2: do both and so forth. And how would you articulate 136 00:06:47,800 --> 00:06:52,120 Speaker 2: based on your career the sort of overlap between the 137 00:06:52,240 --> 00:06:56,600 Speaker 2: skills and techniques you've developed as an arbitrage trader, a 138 00:06:56,680 --> 00:07:00,760 Speaker 2: volatility trader, and the overlap of skills that they're like, Okay, 139 00:07:00,839 --> 00:07:03,800 Speaker 2: this thing that we'll get into that we call soccer analytics. 140 00:07:04,279 --> 00:07:07,279 Speaker 6: Well, soccer is unique because I'd say every aspect of 141 00:07:07,320 --> 00:07:12,200 Speaker 6: the game is a distribution. When you look at on pitch, 142 00:07:12,800 --> 00:07:16,960 Speaker 6: the performance by a team, when you look at performance 143 00:07:17,000 --> 00:07:21,200 Speaker 6: by an individual player, when you look at the seasonal outcomes, 144 00:07:21,440 --> 00:07:26,120 Speaker 6: and this unique component that is promotion relegation. So your 145 00:07:26,200 --> 00:07:31,680 Speaker 6: finances are highly variant year over year. And so there's 146 00:07:31,720 --> 00:07:37,720 Speaker 6: a lot of overlap between volatility trading and working in 147 00:07:37,840 --> 00:07:43,720 Speaker 6: soccer in that you are making highly levered bets on 148 00:07:44,800 --> 00:07:50,120 Speaker 6: often imperfect information. Then that's not necessarily predictive like other 149 00:07:50,160 --> 00:07:51,280 Speaker 6: sports such as baseball. 150 00:07:51,600 --> 00:07:54,280 Speaker 3: Can I ask a very basic question, which is what 151 00:07:54,400 --> 00:07:57,800 Speaker 3: is the point of soccer analytics? So if we go 152 00:07:57,880 --> 00:08:00,520 Speaker 3: back to the market's analogy, we talk about price disco right, 153 00:08:00,600 --> 00:08:02,200 Speaker 3: like you're trying to find the right price for a 154 00:08:02,200 --> 00:08:07,080 Speaker 3: particular asset. With soccer, are you trying to price players, 155 00:08:07,440 --> 00:08:12,160 Speaker 3: improve the training, make more successful predictive bets? I imagine 156 00:08:12,160 --> 00:08:13,400 Speaker 3: it's a bunch of different things. 157 00:08:13,720 --> 00:08:16,080 Speaker 6: Yeah, i'd say everything. I think you said what are 158 00:08:16,120 --> 00:08:18,280 Speaker 6: the things you're trying to do? And I think from 159 00:08:18,440 --> 00:08:22,920 Speaker 6: an investor operator perspective, it's what are you trying to 160 00:08:22,960 --> 00:08:26,240 Speaker 6: not do? You're not trying to get relegated and you're 161 00:08:26,280 --> 00:08:29,840 Speaker 6: not trying to spend thirty million pounds on a player 162 00:08:29,960 --> 00:08:32,280 Speaker 6: that is going to be terrible and you'll have to 163 00:08:32,320 --> 00:08:34,079 Speaker 6: get rid of in two years yours? 164 00:08:34,120 --> 00:08:36,880 Speaker 2: What don't you talk about from your perspective? I mentioned 165 00:08:36,920 --> 00:08:39,960 Speaker 2: your soccer analytics pro actually are a producer. By the way, 166 00:08:40,040 --> 00:08:42,440 Speaker 2: just put in the chat the Bloomberg style guide is football, 167 00:08:43,480 --> 00:08:45,560 Speaker 2: which is injury. Maybe we'll just switch to Yeah, let's 168 00:08:45,559 --> 00:08:49,440 Speaker 2: stay in a company style. You're a football analytics pro 169 00:08:49,960 --> 00:08:52,080 Speaker 2: tracy as like, what is the point? Is it more 170 00:08:52,120 --> 00:08:55,240 Speaker 2: on the team side of things in terms of identifying talent, 171 00:08:55,280 --> 00:08:58,040 Speaker 2: et cetera, or is it more on I don't know, 172 00:08:58,080 --> 00:09:00,800 Speaker 2: I guess like the sort of the best side. What 173 00:09:00,960 --> 00:09:04,120 Speaker 2: is the problem you and your professional capacity are trying to. 174 00:09:04,040 --> 00:09:07,760 Speaker 5: Solve, So it highly depends on the organization. Right, some 175 00:09:08,040 --> 00:09:12,160 Speaker 5: organizations might want to like mic sts, make good hires, 176 00:09:12,320 --> 00:09:16,040 Speaker 5: make sure that they don't lose millions of pounds. Some 177 00:09:16,200 --> 00:09:20,120 Speaker 5: organizations might use soccer analytics to try and improve their strategy, 178 00:09:20,440 --> 00:09:23,679 Speaker 5: but they might look at like in game data and 179 00:09:23,800 --> 00:09:26,400 Speaker 5: try to figure out if there's some optimization that they 180 00:09:26,440 --> 00:09:29,360 Speaker 5: can make based on like where the passes should go 181 00:09:29,520 --> 00:09:32,960 Speaker 5: or where players should be, what play styles make more sense, 182 00:09:33,040 --> 00:09:36,360 Speaker 5: or what yields more expected goals if you will, And 183 00:09:36,400 --> 00:09:38,960 Speaker 5: then I guess the third one is entertainment. There are 184 00:09:38,960 --> 00:09:41,200 Speaker 5: a lot of apps out there, websites out there that 185 00:09:41,440 --> 00:09:46,280 Speaker 5: provide fans with some sort of information, and even back 186 00:09:46,320 --> 00:09:47,960 Speaker 5: in the let's say in the nineties, they would have 187 00:09:47,960 --> 00:09:51,679 Speaker 5: the overlays on the TV where they show number of corners, 188 00:09:52,200 --> 00:09:56,680 Speaker 5: number of yellow cards, and possession percentage. It doesn't mean anything, 189 00:09:56,800 --> 00:09:59,679 Speaker 5: but it's it's interesting. So you can only imagine that 190 00:09:59,720 --> 00:10:01,760 Speaker 5: you can go much deeper with that and people will 191 00:10:01,760 --> 00:10:03,720 Speaker 5: still be engaged. 192 00:10:03,840 --> 00:10:05,680 Speaker 2: Well, if you just say more on tho you said 193 00:10:05,720 --> 00:10:09,600 Speaker 2: it doesn't mean anything, Is that something that has always 194 00:10:09,640 --> 00:10:12,880 Speaker 2: been understood or is this something in twenty twenty six, 195 00:10:13,200 --> 00:10:15,640 Speaker 2: we could say that a few of these stats that 196 00:10:15,679 --> 00:10:18,480 Speaker 2: they like to put on the TV were just extremely 197 00:10:18,600 --> 00:10:20,959 Speaker 2: like low signal data point. 198 00:10:21,080 --> 00:10:23,720 Speaker 5: Yes, so when I say they don't mean anything, I 199 00:10:23,720 --> 00:10:27,000 Speaker 5: don't think those were necessarily predictive of the outcome of 200 00:10:27,040 --> 00:10:30,080 Speaker 5: the match. Got corners might because it's an indicator of 201 00:10:30,679 --> 00:10:33,200 Speaker 5: which team it has the overhand in a game, but 202 00:10:33,280 --> 00:10:37,239 Speaker 5: I'm quite sure most of the others, like possession percentage, 203 00:10:37,400 --> 00:10:40,280 Speaker 5: don't mean all that much. I'm trying to look at 204 00:10:40,480 --> 00:10:41,240 Speaker 5: game outcomes. 205 00:10:41,520 --> 00:10:44,000 Speaker 3: Wait, so we touched on this in the intro. But 206 00:10:44,200 --> 00:10:47,640 Speaker 3: like the perceived wisdom in the sort of early two 207 00:10:47,679 --> 00:10:51,280 Speaker 3: thousands was that soccer was far too complicated to be 208 00:10:51,360 --> 00:10:56,640 Speaker 3: given the baseball moneyball style treatment. What actually changed to 209 00:10:56,679 --> 00:10:59,000 Speaker 3: get us to this point where we're not just talking 210 00:10:59,000 --> 00:11:01,839 Speaker 3: about things like expect goals, but we're also talking about 211 00:11:01,880 --> 00:11:05,600 Speaker 3: like body movements and posturing and things like that. 212 00:11:05,960 --> 00:11:08,120 Speaker 6: You know, baseball had moneyball in two thousand and three, 213 00:11:09,040 --> 00:11:12,160 Speaker 6: and soccer had two things in the twenty ten. So 214 00:11:12,200 --> 00:11:15,760 Speaker 6: in twenty thirteen, Chris Anderson and David Sally published this 215 00:11:15,840 --> 00:11:19,720 Speaker 6: book called The Numbers Game that distilled soccer down to 216 00:11:20,200 --> 00:11:23,880 Speaker 6: more of a weakest length game. And then the other 217 00:11:24,320 --> 00:11:26,840 Speaker 6: thing that happened was I would say we had a 218 00:11:26,840 --> 00:11:31,880 Speaker 6: bit of this revolution on Twitter, which you kind of 219 00:11:31,920 --> 00:11:36,240 Speaker 6: spoke about recently, Joe, where ideas were incubated and it 220 00:11:36,960 --> 00:11:39,480 Speaker 6: I give Michael Kayley a lot of credit for this, 221 00:11:39,600 --> 00:11:42,400 Speaker 6: He's he's been on the pod. But yeah, every Saturday 222 00:11:42,600 --> 00:11:45,360 Speaker 6: in the Premier League you would have six matches and 223 00:11:45,400 --> 00:11:48,120 Speaker 6: then you would wait thirty minutes and Kaylee would just 224 00:11:48,480 --> 00:11:52,560 Speaker 6: post a stream of every expected goals chart and it 225 00:11:52,679 --> 00:11:57,120 Speaker 6: started to stimulate this discussion of does this represent what 226 00:11:57,160 --> 00:12:00,600 Speaker 6: should have happened, what should have been the outcome? I 227 00:12:00,640 --> 00:12:03,680 Speaker 6: think it started to lead us down this path of 228 00:12:04,360 --> 00:12:08,080 Speaker 6: where is x G flawed? It only registers when you 229 00:12:08,160 --> 00:12:11,360 Speaker 6: have a shot, and so there is much more to 230 00:12:11,440 --> 00:12:17,400 Speaker 6: the game than simply which shots occur. There's possession and 231 00:12:18,040 --> 00:12:22,319 Speaker 6: the threat of each possession. There's match momentum, and there's 232 00:12:22,480 --> 00:12:28,160 Speaker 6: changing styles. So it's evolved from on ball data to 233 00:12:28,800 --> 00:12:32,360 Speaker 6: tracking data to now we're going to that deeper layer 234 00:12:32,520 --> 00:12:36,680 Speaker 6: of body pods and what movements can prevent or create 235 00:12:37,040 --> 00:12:38,120 Speaker 6: opportunities to score. 236 00:12:38,440 --> 00:12:42,720 Speaker 5: I think what goes heading out with that is when 237 00:12:42,760 --> 00:12:46,079 Speaker 5: I started ten years ago, soccer football was perceived as 238 00:12:46,200 --> 00:12:48,760 Speaker 5: two complex There were twenty two players, there was a ball. 239 00:12:49,160 --> 00:12:53,600 Speaker 5: People are doing interesting analytics research and applied research already 240 00:12:53,760 --> 00:12:57,480 Speaker 5: in basketball, and what was said as well, there's five 241 00:12:57,520 --> 00:12:59,959 Speaker 5: players on each team, so it's a lot less complex. 242 00:13:00,120 --> 00:13:03,400 Speaker 5: There's more games, and we have more data, and there's 243 00:13:03,400 --> 00:13:07,840 Speaker 5: more scoring, so you could derive metrics a lot easier. 244 00:13:08,360 --> 00:13:09,080 Speaker 6: And I think the. 245 00:13:10,559 --> 00:13:15,400 Speaker 5: Evolution in just how AI was applied in data availability. 246 00:13:15,559 --> 00:13:18,840 Speaker 5: So going from like Mike said, only on ball events 247 00:13:18,880 --> 00:13:21,920 Speaker 5: where we know which player made the pass, which player 248 00:13:21,960 --> 00:13:23,480 Speaker 5: made the shot, but we always have to say, well, 249 00:13:23,520 --> 00:13:25,160 Speaker 5: we don't know where all the other players are because 250 00:13:25,200 --> 00:13:28,640 Speaker 5: those are not recorded. And now we have the ability 251 00:13:28,720 --> 00:13:33,280 Speaker 5: to do make highly complicated artificial intelligent neural nets with 252 00:13:33,440 --> 00:13:35,959 Speaker 5: this positional tracking data where at ten trains per second 253 00:13:36,000 --> 00:13:37,960 Speaker 5: or twenty five points per second, we know where all 254 00:13:37,960 --> 00:13:40,040 Speaker 5: the players are, and we know where the ball is. 255 00:13:40,600 --> 00:13:42,840 Speaker 5: And then additionally, at some point we will get the 256 00:13:42,880 --> 00:13:45,720 Speaker 5: full body posts we know where like the whole skeleton 257 00:13:45,720 --> 00:13:48,560 Speaker 5: of all players are. Basically like that sort of went 258 00:13:48,600 --> 00:13:50,680 Speaker 5: hand in hand and it felt like it was inevitable. 259 00:13:50,720 --> 00:13:53,760 Speaker 5: But also the game is just more complex, so it 260 00:13:53,880 --> 00:14:11,839 Speaker 5: needed more compute, it needed more data, needed more knowledge. 261 00:14:13,880 --> 00:14:17,679 Speaker 2: Tracy Mike mentioned that you know, just looking at a 262 00:14:17,720 --> 00:14:19,480 Speaker 2: shot on goals, so again to tell you so much. 263 00:14:19,480 --> 00:14:22,600 Speaker 2: For example, it doesn't tell you if the referees are 264 00:14:22,600 --> 00:14:25,080 Speaker 2: going to revisit a call made a minute before the 265 00:14:25,120 --> 00:14:28,040 Speaker 2: shot halfway out on the other half of the screen 266 00:14:28,160 --> 00:14:29,320 Speaker 2: and take away the goal. 267 00:14:29,520 --> 00:14:31,200 Speaker 3: This was going to be my question, which we were 268 00:14:31,200 --> 00:14:34,560 Speaker 3: talking about before the podcast recording, But like, how do 269 00:14:34,600 --> 00:14:39,320 Speaker 3: you factor in let's say, the occasional randomness of the 270 00:14:39,400 --> 00:14:43,160 Speaker 3: game and perhaps some erratic trying to be diplomatic here, 271 00:14:43,440 --> 00:14:47,800 Speaker 3: erratic decision making by referees and footballing bodies. 272 00:14:48,480 --> 00:14:53,880 Speaker 6: I think it's control the controllables. That's a fixed parameter 273 00:14:54,240 --> 00:14:58,080 Speaker 6: within the game. Is that uncertainty that you can't control 274 00:14:58,440 --> 00:15:01,600 Speaker 6: and it's hard to predict. There is human error, obviously, 275 00:15:02,320 --> 00:15:07,120 Speaker 6: and it's impossible to isolate when that's going to happen 276 00:15:07,120 --> 00:15:09,240 Speaker 6: in a match, and how it will you just sort 277 00:15:09,280 --> 00:15:09,920 Speaker 6: of play the game. 278 00:15:10,280 --> 00:15:13,200 Speaker 5: You also don't look at individual events per city. You 279 00:15:13,280 --> 00:15:16,920 Speaker 5: might from an analysis perspective, but if you're building a 280 00:15:17,000 --> 00:15:21,120 Speaker 5: basic predictive model for football outcomes, you might just take 281 00:15:21,280 --> 00:15:25,840 Speaker 5: all the scoreboard results from the last ten years and 282 00:15:25,880 --> 00:15:28,320 Speaker 5: then all those things cancel out, Like if one team 283 00:15:28,360 --> 00:15:30,720 Speaker 5: has a red guard somewhere you wouldn't even know from 284 00:15:30,720 --> 00:15:33,600 Speaker 5: these models, but you can build some interesting, let's say 285 00:15:33,640 --> 00:15:37,320 Speaker 5: rudimentary predictive models with just the outcomes. 286 00:15:36,880 --> 00:15:41,200 Speaker 3: Of games out of curiosity. Has var changed football analytics 287 00:15:41,200 --> 00:15:43,600 Speaker 3: at all or presented new opportunities for data. 288 00:15:44,080 --> 00:15:47,640 Speaker 6: It presents new opportunities for data because that is an 289 00:15:47,640 --> 00:15:52,000 Speaker 6: event that happens. If a player is fractionally offside because 290 00:15:52,040 --> 00:15:55,920 Speaker 6: his hand is ahead of the last defender. That doesn't 291 00:15:55,960 --> 00:15:59,360 Speaker 6: take away from the fact that he still got into 292 00:15:59,600 --> 00:16:04,160 Speaker 6: a good opportunity to position, possessed the ball, turned and 293 00:16:04,280 --> 00:16:08,080 Speaker 6: struck the ball in the net. And often that data 294 00:16:08,280 --> 00:16:12,400 Speaker 6: is nullified because of the var But in theory, it's 295 00:16:12,400 --> 00:16:15,160 Speaker 6: something that you should consider in your data set. Within 296 00:16:15,240 --> 00:16:18,840 Speaker 6: the raw data itself, there's tons of data that that 297 00:16:18,880 --> 00:16:24,080 Speaker 6: you could add or sensor out. You know, yours mentioned 298 00:16:24,120 --> 00:16:28,440 Speaker 6: red cards, and a lot of our models we censor 299 00:16:28,560 --> 00:16:32,280 Speaker 6: that out of our data set. There's a there's an 300 00:16:32,360 --> 00:16:38,160 Speaker 6: infamous match from three years ago between Chelsea and Tottenham 301 00:16:38,320 --> 00:16:44,400 Speaker 6: where Tottenham went down to nine men and their response 302 00:16:44,800 --> 00:16:48,800 Speaker 6: was to play a very high line, meaning they put 303 00:16:49,160 --> 00:16:53,560 Speaker 6: all of their defenders up near midfield and tried to 304 00:16:53,600 --> 00:16:58,920 Speaker 6: catch Chelsea off sides, and as one would expect, Chelsea 305 00:16:59,040 --> 00:17:02,760 Speaker 6: proceeded to score or multiple goals later in the match. 306 00:17:02,920 --> 00:17:08,160 Speaker 6: And one player in particular, Nicholas Jackson, had three goals 307 00:17:08,600 --> 00:17:11,359 Speaker 6: in that one match, and when you look at his 308 00:17:12,080 --> 00:17:16,119 Speaker 6: seasonal outputs for that entire season, that three goals was 309 00:17:16,119 --> 00:17:20,120 Speaker 6: probably around twenty percent of his total goals. So when 310 00:17:20,160 --> 00:17:24,720 Speaker 6: something like that happens, when you have an irregular game state, 311 00:17:25,480 --> 00:17:29,679 Speaker 6: it is wise to sort of manipulate your data and 312 00:17:29,800 --> 00:17:32,199 Speaker 6: remove that to give you a full picture of what 313 00:17:32,240 --> 00:17:35,000 Speaker 6: does this game look like at an equal game state. 314 00:17:35,240 --> 00:17:38,000 Speaker 2: Oh, that's interesting. So it's not like that game in 315 00:17:38,040 --> 00:17:42,880 Speaker 2: particular was a rich fountain of data. It's important to 316 00:17:42,960 --> 00:17:46,600 Speaker 2: sort of like recognize that the data from this particular 317 00:17:46,720 --> 00:17:52,560 Speaker 2: game is not going to be particularly predictive about other games. 318 00:17:52,600 --> 00:17:55,040 Speaker 2: And therefore, you a guy who scores three goals in 319 00:17:55,119 --> 00:17:57,480 Speaker 2: that match is probably not going to continue to score 320 00:17:57,880 --> 00:17:59,040 Speaker 2: three goals the game for the rest of it. 321 00:17:59,040 --> 00:18:02,520 Speaker 6: There's a rich fountain of data. Okay, So at the 322 00:18:02,600 --> 00:18:06,119 Speaker 6: end of the season, when you see people analyzing the player, 323 00:18:06,320 --> 00:18:09,639 Speaker 6: they often analyze their season and what do they do 324 00:18:09,720 --> 00:18:13,160 Speaker 6: per ninety minutes of football, And so you have this 325 00:18:13,680 --> 00:18:18,560 Speaker 6: highly skewed data set by some really poor data. Yeah, 326 00:18:18,560 --> 00:18:22,000 Speaker 6: and so there's a lot of data mining involved in 327 00:18:22,040 --> 00:18:26,480 Speaker 6: the process of building a model when you evaluate a 328 00:18:26,760 --> 00:18:29,040 Speaker 6: team and individual players. 329 00:18:29,280 --> 00:18:32,080 Speaker 2: So here's a question I have, and you're talking about 330 00:18:32,160 --> 00:18:35,080 Speaker 2: using neural networks, and it's like, eventually, like you know, 331 00:18:35,119 --> 00:18:38,639 Speaker 2: we'll have the compute to like have the position of 332 00:18:38,720 --> 00:18:43,199 Speaker 2: every player's body flash maybe you know, hundreds of images, 333 00:18:43,520 --> 00:18:46,679 Speaker 2: frames per second and so forth, and then you have 334 00:18:46,760 --> 00:18:50,320 Speaker 2: feeded all into a model, and then we learned something 335 00:18:50,359 --> 00:18:53,200 Speaker 2: about who's more likely to win. But one of the 336 00:18:53,240 --> 00:18:57,199 Speaker 2: things that happens in a lot of other domains, and 337 00:18:57,240 --> 00:19:00,000 Speaker 2: here I'm thinking about like chess or go, for example, 338 00:19:00,520 --> 00:19:03,080 Speaker 2: is that you can have these models that are extraordinary, 339 00:19:03,440 --> 00:19:05,240 Speaker 2: but they don't speak English, or they don't speak any 340 00:19:05,320 --> 00:19:08,480 Speaker 2: human language, and so the transmission of like what was 341 00:19:08,600 --> 00:19:11,840 Speaker 2: learned from these events is something usable by say a 342 00:19:12,000 --> 00:19:15,800 Speaker 2: coach who's thinking about strategy or a general manager who's 343 00:19:15,840 --> 00:19:19,520 Speaker 2: thinking about player selection. Talk to us about, like, when 344 00:19:19,600 --> 00:19:22,680 Speaker 2: you think about these machine learning models that can't really 345 00:19:22,720 --> 00:19:26,760 Speaker 2: communicate their findings in any way in language that humans 346 00:19:26,760 --> 00:19:29,960 Speaker 2: can understand, how you sort of bridge that gap to 347 00:19:30,200 --> 00:19:33,800 Speaker 2: where this is useful information for a team or a manager. 348 00:19:34,160 --> 00:19:37,320 Speaker 5: This is generally be understood I think in the last 349 00:19:37,600 --> 00:19:40,639 Speaker 5: couple of years or the last eight years, as the 350 00:19:40,680 --> 00:19:44,040 Speaker 5: biggest promm you can build these models, these neural nets 351 00:19:44,040 --> 00:19:47,720 Speaker 5: already exist. The main thing is the translation, like you said, 352 00:19:47,800 --> 00:19:52,320 Speaker 5: from model outputs to coach, and I think in the 353 00:19:52,400 --> 00:19:54,520 Speaker 5: last couple of years most most teams have found is 354 00:19:54,560 --> 00:19:57,440 Speaker 5: that they need an expert analyst and data analysts to 355 00:19:57,520 --> 00:20:01,480 Speaker 5: do this conversion or the translations, where the coach isn't 356 00:20:01,520 --> 00:20:04,000 Speaker 5: fed the data directly. Sure the coach is fed just 357 00:20:04,160 --> 00:20:06,960 Speaker 5: the information that the analyst finds from the data, and 358 00:20:07,000 --> 00:20:09,880 Speaker 5: that could be in the end. That generally still boils 359 00:20:09,880 --> 00:20:12,399 Speaker 5: down to having video clips, so you can use the 360 00:20:12,440 --> 00:20:14,600 Speaker 5: data to find video, and then you can show the 361 00:20:14,680 --> 00:20:18,119 Speaker 5: video to the coach, which then from bottom up you 362 00:20:18,200 --> 00:20:18,760 Speaker 5: have this. 363 00:20:19,000 --> 00:20:22,280 Speaker 2: Approach that makes sense. But the part about okay, we're 364 00:20:22,280 --> 00:20:25,000 Speaker 2: going to use the data to find video. So all 365 00:20:25,000 --> 00:20:26,919 Speaker 2: of this makes sense. You have the data special ist 366 00:20:26,920 --> 00:20:30,320 Speaker 2: who translates, you have the video so that there's something tangible. 367 00:20:30,760 --> 00:20:34,000 Speaker 2: But talk to us about that specific step where the 368 00:20:34,119 --> 00:20:37,840 Speaker 2: data analyst sees some sort of model output and then 369 00:20:37,960 --> 00:20:40,840 Speaker 2: is able to use that to find the relevant clip. 370 00:20:41,160 --> 00:20:44,160 Speaker 2: To show the coach something potentially instructive, because that seems 371 00:20:44,200 --> 00:20:45,000 Speaker 2: like the hard part to me. 372 00:20:45,440 --> 00:20:47,919 Speaker 5: Yeah, So the model output could be many things. It 373 00:20:47,920 --> 00:20:51,640 Speaker 5: could be outputs from a classification model that says, in 374 00:20:51,640 --> 00:20:55,000 Speaker 5: this given ten second or fifteen or twenty second window, 375 00:20:55,640 --> 00:20:57,760 Speaker 5: this team played in this sort of build up, or 376 00:20:57,920 --> 00:21:00,480 Speaker 5: they have this type of structure. It could also be 377 00:21:00,800 --> 00:21:03,160 Speaker 5: a little bit more advanced where you can simply say, well, 378 00:21:03,200 --> 00:21:05,840 Speaker 5: in this instance, we had a high probability of conceding 379 00:21:05,880 --> 00:21:08,160 Speaker 5: a goal. And that could be just from an expected 380 00:21:08,200 --> 00:21:10,320 Speaker 5: goal shot if you're looking at only a cent data. 381 00:21:10,359 --> 00:21:13,199 Speaker 5: But you could also have model outputs from something we 382 00:21:13,280 --> 00:21:16,240 Speaker 5: call an expected possession value model or an expected tread model, 383 00:21:16,240 --> 00:21:18,760 Speaker 5: where you measure the chance that the teams going to 384 00:21:18,760 --> 00:21:21,560 Speaker 5: score and let's say the next thirty seconds or the 385 00:21:21,560 --> 00:21:24,280 Speaker 5: next possession, and then you can find this fight there 386 00:21:24,680 --> 00:21:27,760 Speaker 5: and either measure when your team is likely to concede 387 00:21:27,920 --> 00:21:29,920 Speaker 5: or measure when the other team is likely to score, 388 00:21:30,080 --> 00:21:32,560 Speaker 5: and you can use those kind of signals to boil 389 00:21:32,600 --> 00:21:33,320 Speaker 5: it down to video. 390 00:21:33,680 --> 00:21:35,719 Speaker 3: Yeah, this is something I wanted to ask. Actually, so 391 00:21:36,080 --> 00:21:39,399 Speaker 3: you mentioned speed just then, like, what is the actual 392 00:21:39,520 --> 00:21:42,120 Speaker 3: latency that we're talking about since we're using all these 393 00:21:42,119 --> 00:21:46,280 Speaker 3: market Yeah, jeez, Like, are we talking about an insight 394 00:21:46,440 --> 00:21:50,439 Speaker 3: that's actionable within seconds, like you're going to sub a 395 00:21:50,480 --> 00:21:53,960 Speaker 3: player on after your model splits something out like live 396 00:21:54,080 --> 00:21:57,600 Speaker 3: during the game. Or is it more realistically that you're 397 00:21:57,680 --> 00:22:00,680 Speaker 3: reviewing the model and the analytics a d a game 398 00:22:00,880 --> 00:22:04,520 Speaker 3: and sort of tweaking. I guess when things have calmed down. 399 00:22:05,119 --> 00:22:08,040 Speaker 5: Apparently most of this does not happen live, So most 400 00:22:08,040 --> 00:22:11,040 Speaker 5: of it happens either pre match or postmatch. 401 00:22:11,080 --> 00:22:12,399 Speaker 6: But you can still do it. 402 00:22:12,400 --> 00:22:16,080 Speaker 5: There's there's enough live data to make these instances the worthwhile. 403 00:22:16,440 --> 00:22:19,240 Speaker 6: Yeah, I'd say most those types of adjustments based on 404 00:22:19,440 --> 00:22:23,240 Speaker 6: data typically happen at halftime or if you're in the 405 00:22:23,240 --> 00:22:27,560 Speaker 6: World Cup, during a higher dration break. And in the 406 00:22:27,600 --> 00:22:33,560 Speaker 6: derivatives world, I always think of expected outcome versus realized outcome. 407 00:22:34,160 --> 00:22:36,840 Speaker 6: So what is the market implying? What do you expect? 408 00:22:36,840 --> 00:22:41,360 Speaker 6: And then what is actually happening? And that is sort 409 00:22:41,359 --> 00:22:45,000 Speaker 6: of the nexus of how teams prepare for matches. They 410 00:22:45,000 --> 00:22:48,040 Speaker 6: come up with their own expected outcome, How is our 411 00:22:48,080 --> 00:22:51,399 Speaker 6: opposition going to play? How are we going to play? 412 00:22:52,240 --> 00:22:57,760 Speaker 6: And then that live in game is your realized outcome? 413 00:22:58,000 --> 00:23:01,720 Speaker 6: And so you're receiving that data and it's being transmitted 414 00:23:01,720 --> 00:23:05,000 Speaker 6: to analysts who can then communicate it down to the 415 00:23:05,040 --> 00:23:08,440 Speaker 6: bench to discuss with the manager, who can then make 416 00:23:08,480 --> 00:23:10,280 Speaker 6: those changes at that halftime. 417 00:23:10,680 --> 00:23:13,639 Speaker 3: It's interesting the game of two halves, Joe's now the 418 00:23:13,680 --> 00:23:17,480 Speaker 3: game of four quarters, offering up more opportunities to make 419 00:23:17,520 --> 00:23:18,639 Speaker 3: model based adjustments. 420 00:23:18,680 --> 00:23:19,600 Speaker 4: That's right, we thought about it. 421 00:23:19,720 --> 00:23:21,720 Speaker 2: We have We used to have two discrete events in 422 00:23:21,760 --> 00:23:23,600 Speaker 2: a game, the first half of a second, and now 423 00:23:23,640 --> 00:23:26,240 Speaker 2: we have at least four. So I guess that, yeah, 424 00:23:26,320 --> 00:23:29,640 Speaker 2: that that creates more data as well as maybe more 425 00:23:29,640 --> 00:23:35,320 Speaker 2: adjustment as well as more ad revenue. You know, obviously 426 00:23:35,760 --> 00:23:40,840 Speaker 2: in baseball at least according to Michael lewis right that 427 00:23:40,880 --> 00:23:43,679 Speaker 2: there was this period where you know, you had the 428 00:23:43,720 --> 00:23:46,520 Speaker 2: old time scouts and they're like, oh, this guy has 429 00:23:46,560 --> 00:23:48,720 Speaker 2: good hustle, right, or this guy has a good heart. 430 00:23:48,760 --> 00:23:50,919 Speaker 2: There was like no data behind any of it. He 431 00:23:51,000 --> 00:23:53,760 Speaker 2: may just sort of had the you know, the swagger 432 00:23:53,840 --> 00:23:57,119 Speaker 2: of someone who looked like maybe a star player, and 433 00:23:57,160 --> 00:23:58,840 Speaker 2: then it's like, oh no, but he look at his 434 00:23:59,000 --> 00:24:02,560 Speaker 2: like on base person, that's his vorp or whatever, and 435 00:24:02,600 --> 00:24:05,240 Speaker 2: then they get promoted. But it was like a culture 436 00:24:05,320 --> 00:24:09,800 Speaker 2: thing has there been a similar cultural clash within sort 437 00:24:09,840 --> 00:24:13,680 Speaker 2: of soccer scouting, where what the data says about what 438 00:24:13,800 --> 00:24:18,159 Speaker 2: constitutes a player does not map to traditional intuitions. 439 00:24:18,640 --> 00:24:20,240 Speaker 6: I think a lot of clubs look at it from 440 00:24:20,560 --> 00:24:24,639 Speaker 6: two perspectives. I don't think there's this old school scouts 441 00:24:24,800 --> 00:24:27,520 Speaker 6: versus the data guy of mentality. I think it's a 442 00:24:27,600 --> 00:24:29,840 Speaker 6: very collaborative. I think what a lot you know, what 443 00:24:29,880 --> 00:24:32,520 Speaker 6: a lot of clubs do is they have an individual 444 00:24:32,560 --> 00:24:35,399 Speaker 6: scout who will go watch a player and give his 445 00:24:35,480 --> 00:24:39,560 Speaker 6: assessment in his rating, and then they will have their 446 00:24:39,600 --> 00:24:45,120 Speaker 6: own internal model with data, and they will look at 447 00:24:45,160 --> 00:24:49,639 Speaker 6: the delta's between those two different models, and if something 448 00:24:49,720 --> 00:24:54,719 Speaker 6: seems off, you often have collaboration between the data person 449 00:24:54,960 --> 00:25:00,439 Speaker 6: and the scout and they figure out who's right, who's wrong. 450 00:25:00,960 --> 00:25:04,720 Speaker 6: And then to your point about hustle, let's call it. Yeah, 451 00:25:04,800 --> 00:25:09,879 Speaker 6: there are Swiss army knife ways to quantify that in 452 00:25:10,440 --> 00:25:14,960 Speaker 6: soccer in certain instances, if you look at game state, 453 00:25:15,119 --> 00:25:18,080 Speaker 6: say the game state. You know when game state is 454 00:25:18,200 --> 00:25:23,800 Speaker 6: essentially zero plus one minus one plus two minus two 455 00:25:24,000 --> 00:25:27,160 Speaker 6: plus three minus three, and so say you're in a 456 00:25:27,200 --> 00:25:30,840 Speaker 6: plus three minus three game state, and win probability for 457 00:25:31,440 --> 00:25:35,960 Speaker 6: one team is ninety eight percent. You can manipulate the 458 00:25:36,040 --> 00:25:41,400 Speaker 6: data and only look at how players are performing in 459 00:25:41,440 --> 00:25:44,280 Speaker 6: that game. State you're out of the game, are you 460 00:25:44,359 --> 00:25:46,480 Speaker 6: still competing? Do you still care? Are you in the 461 00:25:46,560 --> 00:25:49,960 Speaker 6: right position now? That may not align with the old 462 00:25:49,960 --> 00:25:53,760 Speaker 6: school scout saying, you know, this guy has grit, But 463 00:25:54,600 --> 00:25:58,439 Speaker 6: again there's there's Swiss army knife ways to give some 464 00:25:58,520 --> 00:26:02,359 Speaker 6: type of indication, you know, to say, soccer data creates questions, 465 00:26:02,440 --> 00:26:03,840 Speaker 6: It doesn't give us answers. 466 00:26:04,119 --> 00:26:07,199 Speaker 3: Wait, just to better understand this, can you give us 467 00:26:07,240 --> 00:26:11,200 Speaker 3: like an analytics framework if you were trying to judge 468 00:26:11,680 --> 00:26:13,880 Speaker 3: the best This is the loaded question, but it comes 469 00:26:13,920 --> 00:26:16,000 Speaker 3: up on every discussion. But you're trying to judge the 470 00:26:16,000 --> 00:26:19,639 Speaker 3: best soccer player, either of all time or currently, Like, 471 00:26:19,680 --> 00:26:22,760 Speaker 3: what would the analytics framework for that actually look like? 472 00:26:22,960 --> 00:26:24,200 Speaker 6: Jude Bellingham? 473 00:26:24,640 --> 00:26:28,639 Speaker 2: Okay, yeah, yeah, Like really seriously, this is a great question, Like, 474 00:26:28,720 --> 00:26:31,560 Speaker 2: walk us through what the math says about Jude Bellingham 475 00:26:31,560 --> 00:26:32,760 Speaker 2: and how you would derive that. 476 00:26:32,880 --> 00:26:38,160 Speaker 6: Well, Jude Bellingham can play four or five different positions. Right, 477 00:26:38,320 --> 00:26:41,560 Speaker 6: most players they have the number nine and their role 478 00:26:41,720 --> 00:26:45,960 Speaker 6: is number nine. But if you think of the game 479 00:26:46,080 --> 00:26:50,520 Speaker 6: as this book with different chapters within the story, Jude 480 00:26:50,520 --> 00:26:56,440 Speaker 6: Bellingham can perform whatever task he needs to perform at 481 00:26:56,480 --> 00:26:59,360 Speaker 6: every single chapter throughout the book, and he does it 482 00:26:59,720 --> 00:27:02,960 Speaker 6: at the highest level. He could play any position on 483 00:27:03,000 --> 00:27:04,719 Speaker 6: the pitch besides goalkeeper. 484 00:27:04,920 --> 00:27:08,199 Speaker 2: And I just started like, how is this established? Like 485 00:27:08,680 --> 00:27:10,879 Speaker 2: someone could say, oh, this guy, they say this is 486 00:27:10,960 --> 00:27:13,399 Speaker 2: baseball too, he's a good all around player. Whatever we 487 00:27:13,400 --> 00:27:16,600 Speaker 2: can see, But like, what is the data that actually 488 00:27:16,720 --> 00:27:22,040 Speaker 2: establishes that Jude Bellingham can play at high levels in 489 00:27:22,080 --> 00:27:24,720 Speaker 2: a wide range of position. How do you derive that 490 00:27:24,880 --> 00:27:26,240 Speaker 2: conclusion quantitatively? 491 00:27:26,680 --> 00:27:28,879 Speaker 6: So from a data lens, we think of it in 492 00:27:29,000 --> 00:27:33,119 Speaker 6: two fronts, in possession and out of possession. Okay, So 493 00:27:33,640 --> 00:27:36,520 Speaker 6: in possession is how you're progressing the ball into threatening 494 00:27:36,560 --> 00:27:40,040 Speaker 6: areas and obviously creating high probability opportunities. Then out of 495 00:27:40,119 --> 00:27:46,639 Speaker 6: possession is how are you preventing a team from moving 496 00:27:46,680 --> 00:27:49,399 Speaker 6: the ball into threatening areas. Since it's a very tricky 497 00:27:49,480 --> 00:27:52,879 Speaker 6: question sports, because you're trying to quantify the value of 498 00:27:52,920 --> 00:27:57,720 Speaker 6: an event that does not happen. So let's say Joe, 499 00:27:57,760 --> 00:28:00,200 Speaker 6: you have the ball out on the wing, Tracy, you're 500 00:28:00,240 --> 00:28:03,359 Speaker 6: right in front of the box. Jude Bellingham, he would 501 00:28:03,359 --> 00:28:06,040 Speaker 6: be both. He was always moving into that passing lane 502 00:28:06,240 --> 00:28:08,439 Speaker 6: is right in between you guys at the right moment, 503 00:28:09,640 --> 00:28:14,240 Speaker 6: and we have the ability to quantify the value of 504 00:28:14,760 --> 00:28:19,879 Speaker 6: those movements and the closure of these lanes as players 505 00:28:19,920 --> 00:28:22,199 Speaker 6: move out on the pitch, and then you'll obviously get 506 00:28:22,240 --> 00:28:24,320 Speaker 6: to the next phase where it's how do they do 507 00:28:24,400 --> 00:28:25,280 Speaker 6: this with their feet? 508 00:28:41,280 --> 00:28:44,600 Speaker 2: We started talking about this sort of translation from what 509 00:28:44,680 --> 00:28:47,360 Speaker 2: the model says to a coach or something like that, 510 00:28:47,480 --> 00:28:50,920 Speaker 2: and that still seem as tricky for better someone who's 511 00:28:50,920 --> 00:28:53,600 Speaker 2: betting on sports, that might be totally irrelevant, like they're 512 00:28:53,680 --> 00:28:56,160 Speaker 2: just like here, the model says this, this team is better. 513 00:28:56,680 --> 00:28:58,520 Speaker 2: The line doesn't match up with this. They're for going 514 00:28:58,560 --> 00:28:59,880 Speaker 2: to bet on this team. I don't know why them 515 00:28:59,880 --> 00:29:02,200 Speaker 2: a model says, My model says this team is better, 516 00:29:02,200 --> 00:29:04,760 Speaker 2: but it does. So the translation is unnecessary if you're 517 00:29:04,800 --> 00:29:07,360 Speaker 2: just betting, like are we at the stage or are 518 00:29:07,360 --> 00:29:10,440 Speaker 2: we getting close to the stage where you could, for example, 519 00:29:11,120 --> 00:29:14,000 Speaker 2: feed a model the first five minutes of a game 520 00:29:14,240 --> 00:29:17,280 Speaker 2: scoreless zero zero, and to all the players like, oh, 521 00:29:17,320 --> 00:29:19,400 Speaker 2: this looks this looks like a competitive match. It's going 522 00:29:19,480 --> 00:29:22,239 Speaker 2: to be good both sides. But models are able to 523 00:29:22,320 --> 00:29:26,360 Speaker 2: detect something that we can't articulate that says, oh no, 524 00:29:26,480 --> 00:29:28,800 Speaker 2: this team is playing and even though it looks like 525 00:29:28,800 --> 00:29:32,120 Speaker 2: a tie game and a competitive one, actually for reasons 526 00:29:32,120 --> 00:29:35,120 Speaker 2: that we can't put into english, put into language, this 527 00:29:35,200 --> 00:29:37,160 Speaker 2: team looks like they're going to win the game seven. 528 00:29:37,240 --> 00:29:40,200 Speaker 2: They have a seventy percent chance. Is that a thing 529 00:29:40,360 --> 00:29:42,520 Speaker 2: or is that a phenomenon or is that a realistic 530 00:29:42,560 --> 00:29:43,360 Speaker 2: thing to expect? 531 00:29:43,600 --> 00:29:47,400 Speaker 5: And there are game with probability models, yeah, which start 532 00:29:47,480 --> 00:29:50,560 Speaker 5: with just a pre game cheap strength, so both teamch 533 00:29:50,600 --> 00:29:54,640 Speaker 5: have some value. Perhaps you can think of an ELO rating, okay, 534 00:29:55,320 --> 00:29:58,200 Speaker 5: and they boiled onto a win, a draw, and a 535 00:29:58,240 --> 00:30:02,200 Speaker 5: lost percentage, and then those percentages can in game be updated, 536 00:30:02,200 --> 00:30:04,280 Speaker 5: but they won't swing all that much because you have 537 00:30:04,360 --> 00:30:08,880 Speaker 5: this prior information. So maybe after the first minute, one 538 00:30:08,920 --> 00:30:12,000 Speaker 5: team has some egg, or maybe after the first five minutes, 539 00:30:12,080 --> 00:30:15,800 Speaker 5: let's say some team has created some high probability chances 540 00:30:15,800 --> 00:30:18,600 Speaker 5: and the team might be the underdog team. This was 541 00:30:18,640 --> 00:30:21,880 Speaker 5: highly unexpected, I guess since since the ELW rating set 542 00:30:21,880 --> 00:30:24,400 Speaker 5: that the other team would would be he the over end. 543 00:30:24,440 --> 00:30:26,840 Speaker 5: So you can you can slightly update your beliefs there. 544 00:30:27,200 --> 00:30:30,600 Speaker 5: I don't think it's like changing the needle or moving 545 00:30:30,600 --> 00:30:32,880 Speaker 5: the needle all that much. Given that it's just five 546 00:30:32,880 --> 00:30:35,840 Speaker 5: minutes of information, but you can definitely update your beliefs 547 00:30:35,880 --> 00:30:40,440 Speaker 5: throughout the game given chances created, expected possession value as 548 00:30:40,480 --> 00:30:43,080 Speaker 5: we just talked about, or momentum or something in between, 549 00:30:43,160 --> 00:30:45,200 Speaker 5: depending on the data you have available to you. 550 00:30:46,400 --> 00:30:48,680 Speaker 3: Mike, I wanted to ask you, given your involvement with 551 00:30:48,880 --> 00:30:53,040 Speaker 3: Austin FC, we know that there are obviously differences between 552 00:30:53,600 --> 00:30:58,000 Speaker 3: Major League Soccer and European leagues, and some of those are, 553 00:30:58,280 --> 00:31:02,120 Speaker 3: you know, things like salary caps, designated players, roster rules, 554 00:31:02,840 --> 00:31:05,600 Speaker 3: lack of relegation, lack of Yeah, that's a big one. 555 00:31:06,280 --> 00:31:10,400 Speaker 3: Does that actually make soccer analytics and MLS like a 556 00:31:10,400 --> 00:31:13,640 Speaker 3: little bit cleaner in some ways? In the sense that, 557 00:31:13,800 --> 00:31:17,440 Speaker 3: like in the European leagues, the money is again trying 558 00:31:17,440 --> 00:31:20,560 Speaker 3: to be diplomatic here, but it's very free flowing. There's 559 00:31:20,600 --> 00:31:23,720 Speaker 3: a bit of rule stretching going on at times when 560 00:31:23,720 --> 00:31:27,120 Speaker 3: it comes to salary restrictions and things like that. Like 561 00:31:27,480 --> 00:31:31,920 Speaker 3: compare MLS analytics versus European analytics for US. 562 00:31:32,240 --> 00:31:36,320 Speaker 6: So, in European analytics you have essentially your recruitment and 563 00:31:36,360 --> 00:31:41,160 Speaker 6: then first team analysis. In MLS you have recruitment first 564 00:31:41,200 --> 00:31:43,680 Speaker 6: team analysis, but you have this third vector that I 565 00:31:43,880 --> 00:31:49,120 Speaker 6: call portfolio management. Right, this cap structure in the MLS 566 00:31:49,560 --> 00:31:53,760 Speaker 6: is put in place with the intention of creating parity, 567 00:31:53,960 --> 00:31:58,040 Speaker 6: and as you mentioned, it's highly complicated. The best way 568 00:31:58,680 --> 00:32:01,920 Speaker 6: for you to understand it would be like me saying, Tracy, 569 00:32:02,000 --> 00:32:05,240 Speaker 6: I'm going to give you ten million dollars. You can 570 00:32:05,320 --> 00:32:11,240 Speaker 6: spend two million dollars on Navidia, seven million dollars on Walmart, 571 00:32:11,320 --> 00:32:15,520 Speaker 6: and one million dollars on a speculative biotech stock. And 572 00:32:15,640 --> 00:32:19,479 Speaker 6: so you have to think of each player from a 573 00:32:20,160 --> 00:32:24,520 Speaker 6: relative value perspective based on where they slot in your 574 00:32:24,560 --> 00:32:25,280 Speaker 6: cap structure. 575 00:32:26,000 --> 00:32:26,880 Speaker 4: Has that worked? 576 00:32:27,040 --> 00:32:29,680 Speaker 2: I mean with MLS. So it's like, I understand, you 577 00:32:29,720 --> 00:32:32,280 Speaker 2: have this new league that you know, you don't want 578 00:32:32,360 --> 00:32:34,920 Speaker 2: some really rich team to win all the time, and 579 00:32:34,960 --> 00:32:38,560 Speaker 2: then I don't know, only Miami or wins, and then 580 00:32:38,720 --> 00:32:41,280 Speaker 2: fans lose interest in the rest of the country. I'm 581 00:32:41,320 --> 00:32:43,160 Speaker 2: just I don't know what the actual I think my 582 00:32:43,280 --> 00:32:46,080 Speaker 2: understanding is Austin in particular is a very big fan base. 583 00:32:46,400 --> 00:32:50,040 Speaker 2: But has that worked in practice to sort of create 584 00:32:50,080 --> 00:32:54,400 Speaker 2: an equal level of like fan affinity that's geographically distributed. 585 00:32:54,520 --> 00:32:57,000 Speaker 6: Well at Austin We're we're eling nine months into our 586 00:32:57,040 --> 00:33:00,440 Speaker 6: project here, and we just had some turn and over 587 00:33:00,880 --> 00:33:05,080 Speaker 6: with our sporting department, and so I would say that 588 00:33:05,200 --> 00:33:10,240 Speaker 6: we have yet to integrate and prove this portfolio management 589 00:33:10,880 --> 00:33:15,479 Speaker 6: theory and the impact on success in points in the table. 590 00:33:16,160 --> 00:33:19,480 Speaker 6: But as a market participant, when I read these rules, 591 00:33:19,720 --> 00:33:23,200 Speaker 6: I like, my brain immediately goes to the markets and 592 00:33:23,920 --> 00:33:24,880 Speaker 6: portfolio allocation. 593 00:33:25,080 --> 00:33:27,640 Speaker 3: Okay, so we have to watch Austin FC as a 594 00:33:27,680 --> 00:33:31,720 Speaker 3: test case for the portfolio management thesis in football. 595 00:33:31,320 --> 00:33:35,120 Speaker 2: Boy, Just to be clear, so this portfolio management thesis, 596 00:33:35,160 --> 00:33:38,680 Speaker 2: would you obviously have a strong intuition for having traded 597 00:33:38,760 --> 00:33:42,760 Speaker 2: volatility for a long time. This is the framework that 598 00:33:42,880 --> 00:33:46,600 Speaker 2: you're bringing to your work in Austin. 599 00:33:46,920 --> 00:33:51,920 Speaker 6: Yeah. Correct. Every player has a designated slot. So a 600 00:33:51,960 --> 00:33:56,720 Speaker 6: player that could come in as a DP could be terrible. 601 00:33:57,160 --> 00:33:58,040 Speaker 6: But if he's going to. 602 00:33:58,080 --> 00:34:02,000 Speaker 4: Do oh yeah, okay. 603 00:34:02,400 --> 00:34:04,680 Speaker 6: But that same player, if he is going to be 604 00:34:05,800 --> 00:34:12,200 Speaker 6: a senior minimum salary player, could be in the top 605 00:34:12,360 --> 00:34:16,840 Speaker 6: percentile of talent for that specific slot. And then within 606 00:34:16,960 --> 00:34:22,560 Speaker 6: the league itself, you know, you have different roster construction strategies. 607 00:34:22,680 --> 00:34:27,200 Speaker 6: You can either have three designated players or you could 608 00:34:27,360 --> 00:34:31,680 Speaker 6: opt for two designated players and for U twenty two 609 00:34:31,760 --> 00:34:36,239 Speaker 6: players and The way the league works is they have 610 00:34:36,520 --> 00:34:40,560 Speaker 6: a They have a salary cap, but it's more of 611 00:34:40,600 --> 00:34:45,600 Speaker 6: a salary cap charge. So whilst Lionel Messi may be 612 00:34:45,760 --> 00:34:50,359 Speaker 6: making over twenty million dollars in salary, his salary cap 613 00:34:50,480 --> 00:34:54,879 Speaker 6: charge as a designated player is going to be much 614 00:34:54,920 --> 00:34:57,360 Speaker 6: lower than that. It could be seven hundred and fifty thousand. 615 00:34:57,520 --> 00:34:58,560 Speaker 2: Wait, I don't understand that. 616 00:34:58,640 --> 00:35:03,200 Speaker 6: Yeah, it's confusing. There's a salary cap charge based on 617 00:35:03,320 --> 00:35:07,560 Speaker 6: these roster designations, so every slot has a dollar charge 618 00:35:07,600 --> 00:35:11,560 Speaker 6: associated with it. So you have to work within that 619 00:35:11,680 --> 00:35:15,360 Speaker 6: framework of what is the charge for each player and 620 00:35:15,960 --> 00:35:18,920 Speaker 6: your finite amount of kapala you're allowed to spet. 621 00:35:19,239 --> 00:35:22,319 Speaker 3: So the designated players are the ones you're allowed to 622 00:35:22,400 --> 00:35:22,959 Speaker 3: pay more. 623 00:35:23,400 --> 00:35:23,839 Speaker 5: Got it? 624 00:35:24,040 --> 00:35:27,640 Speaker 3: Because this is the legacy of like La Galaxy getting 625 00:35:27,719 --> 00:35:30,480 Speaker 3: David Beckham and wanting to pay him right, lots and 626 00:35:30,520 --> 00:35:32,280 Speaker 3: lots and lots and lots of money. 627 00:35:32,600 --> 00:35:33,760 Speaker 4: This is helpful, Okay. 628 00:35:33,880 --> 00:35:36,839 Speaker 3: So one of the criticisms of modern football, I guess, 629 00:35:36,960 --> 00:35:42,000 Speaker 3: is that it's dominated by the wealthiest clubs, like whoever 630 00:35:42,040 --> 00:35:45,040 Speaker 3: has the most money can buy the best players, certainly 631 00:35:45,320 --> 00:35:48,440 Speaker 3: in Europe, and so the big just kind of get bigger, 632 00:35:49,080 --> 00:35:52,719 Speaker 3: And I could certainly see an argument where if analytics 633 00:35:52,760 --> 00:35:55,920 Speaker 3: becomes more important to actually playing the game, then whoever 634 00:35:55,960 --> 00:35:59,319 Speaker 3: has the most resources, the most compute, the most engineers 635 00:35:59,360 --> 00:36:02,400 Speaker 3: I guess nowadays is going to have an edge here. 636 00:36:02,800 --> 00:36:06,000 Speaker 3: But on the other hand, we have had technology before 637 00:36:06,080 --> 00:36:09,160 Speaker 3: that has a sort of democratizing effect. 638 00:36:09,239 --> 00:36:10,680 Speaker 1: Yeah, we have open source. 639 00:36:10,440 --> 00:36:13,640 Speaker 3: Models, all of that, and I think there have been 640 00:36:13,719 --> 00:36:17,040 Speaker 3: some instances of the smaller clubs actually like using this 641 00:36:17,160 --> 00:36:20,319 Speaker 3: technology to perform better. But which way are we going 642 00:36:20,400 --> 00:36:22,720 Speaker 3: to go? Is it the big get bigger or maybe 643 00:36:22,760 --> 00:36:25,640 Speaker 3: we see some smaller clubs level the playing field here. 644 00:36:26,360 --> 00:36:27,359 Speaker 6: I hope it's the latter. 645 00:36:27,640 --> 00:36:31,160 Speaker 5: Yeah, same, So you mentioned open source models actually build 646 00:36:31,800 --> 00:36:34,799 Speaker 5: open source software to help these smaller clubs. I guess 647 00:36:34,800 --> 00:36:37,839 Speaker 5: it's also helping the bigger clubs, but it allows them 648 00:36:37,960 --> 00:36:42,080 Speaker 5: to build these crossnural nets and these expected possession telling models, 649 00:36:42,480 --> 00:36:46,439 Speaker 5: and then also load these tracting data districting data, which 650 00:36:46,480 --> 00:36:49,640 Speaker 5: has been a big task just in general because of 651 00:36:49,719 --> 00:36:53,280 Speaker 5: the data structure and the data size, and the software 652 00:36:53,280 --> 00:36:56,440 Speaker 5: that I that I work on helps clubs any club 653 00:36:56,600 --> 00:36:59,879 Speaker 5: basically get get started. So I hope by doing that 654 00:37:00,080 --> 00:37:03,960 Speaker 5: it will help the smaller clubs, help data providers I 655 00:37:03,960 --> 00:37:08,239 Speaker 5: guess provide better insights to level the playing fields. But 656 00:37:08,520 --> 00:37:11,360 Speaker 5: I'm not sure. I'm not sure that it is working 657 00:37:11,360 --> 00:37:15,080 Speaker 5: out necessarily because getting the knowledge, like we talked about before, 658 00:37:15,080 --> 00:37:18,360 Speaker 5: actually I'm distilling the knowledge from the data. I still 659 00:37:19,040 --> 00:37:19,440 Speaker 5: I think. 660 00:37:19,960 --> 00:37:20,640 Speaker 6: One of the problem. 661 00:37:20,680 --> 00:37:23,000 Speaker 3: Next. Oh yeah, so this reminds me. I wanted to 662 00:37:23,000 --> 00:37:25,800 Speaker 3: ask as well, who actually owns the data here? Where's 663 00:37:25,840 --> 00:37:26,640 Speaker 3: the data come from? 664 00:37:26,760 --> 00:37:29,200 Speaker 5: It depends, okay, and I think some of it is 665 00:37:29,239 --> 00:37:31,239 Speaker 5: a gray area. There are a lot of data providers, 666 00:37:31,680 --> 00:37:35,160 Speaker 5: some have licenses, some don't. Yeah, it's a big question 667 00:37:35,239 --> 00:37:37,000 Speaker 5: market in some instances. 668 00:37:37,840 --> 00:37:41,480 Speaker 6: I'll tell you a story about when I first started 669 00:37:41,520 --> 00:37:45,960 Speaker 6: building models for AFC Bourne, myth back in the Premier 670 00:37:46,000 --> 00:37:50,600 Speaker 6: League in twenty sixteen. I had no idea where I 671 00:37:50,600 --> 00:37:54,759 Speaker 6: could find the data, and so I went on upwork 672 00:37:55,280 --> 00:37:58,839 Speaker 6: and I posted an ad and I just said I 673 00:37:58,880 --> 00:38:04,960 Speaker 6: need a developer who has worked with European football betters, 674 00:38:05,000 --> 00:38:09,160 Speaker 6: and a gentleman named Dmitri and Ukraine replied to me 675 00:38:09,200 --> 00:38:12,600 Speaker 6: and he said, yeah, I've worked with many professional betters. 676 00:38:12,960 --> 00:38:14,960 Speaker 6: And I said, give me all the data that you 677 00:38:14,960 --> 00:38:17,560 Speaker 6: can find. And so it was very skunk works, but 678 00:38:18,080 --> 00:38:21,480 Speaker 6: Dmitri was able to find a way for me to 679 00:38:21,520 --> 00:38:26,000 Speaker 6: get access to all of the on ball data across 680 00:38:26,040 --> 00:38:29,320 Speaker 6: the world at the time to start building my models. 681 00:38:29,600 --> 00:38:31,520 Speaker 2: What is the data? So it's like, okay, you find 682 00:38:31,560 --> 00:38:36,960 Speaker 2: some data provider or Dmitri and Ukraine collect data. Is 683 00:38:37,000 --> 00:38:41,279 Speaker 2: this numerical data? Is this a series of frames? Like 684 00:38:41,360 --> 00:38:42,960 Speaker 2: what can you when you say okay, you need to 685 00:38:42,960 --> 00:38:45,600 Speaker 2: go out and get the data. What form are you 686 00:38:45,719 --> 00:38:46,160 Speaker 2: getting it? 687 00:38:46,200 --> 00:38:47,120 Speaker 1: What does that mean? 688 00:38:47,520 --> 00:38:47,640 Speaker 6: Like? 689 00:38:47,680 --> 00:38:48,480 Speaker 2: What does it look like? 690 00:38:48,719 --> 00:38:54,480 Speaker 6: So it's varied throughout the years, but the most prevalent 691 00:38:54,600 --> 00:38:59,319 Speaker 6: generic data out there is on ball data. And on 692 00:38:59,480 --> 00:39:02,279 Speaker 6: ball data if you're just put on your Excel hat 693 00:39:02,400 --> 00:39:05,560 Speaker 6: and think about what it looks like in an Excel spreadsheet, 694 00:39:05,880 --> 00:39:10,960 Speaker 6: has every single event tagged, so you have it just 695 00:39:11,040 --> 00:39:13,920 Speaker 6: look goes down a series of events past past, past 696 00:39:14,000 --> 00:39:19,960 Speaker 6: drible shot, goal, past past drible tackle. It has the event, 697 00:39:20,680 --> 00:39:25,759 Speaker 6: it has the player or players involved, It has the 698 00:39:26,000 --> 00:39:30,160 Speaker 6: X Y coordinate on a pitch, and it has the time. 699 00:39:30,520 --> 00:39:35,799 Speaker 6: And so when you have these raw data points, you 700 00:39:35,840 --> 00:39:42,200 Speaker 6: can conditionally put together a mosaic of what is happening 701 00:39:42,320 --> 00:39:47,360 Speaker 6: on the pitch because you can measure the events, the speed, 702 00:39:47,920 --> 00:39:52,640 Speaker 6: and the location. Tracking data is a bit more nuanced. 703 00:39:52,960 --> 00:39:55,120 Speaker 6: I'll let yours touch on that. 704 00:39:55,480 --> 00:39:58,480 Speaker 5: So you can still imagine it an Excel spreadsheet, but 705 00:39:58,480 --> 00:40:00,600 Speaker 5: I don't think your Excel spreadsheet would like it very 706 00:40:00,640 --> 00:40:03,359 Speaker 5: much if you try to load in this data. It's 707 00:40:03,400 --> 00:40:08,200 Speaker 5: basically a player identify a team identifier, and then X 708 00:40:08,239 --> 00:40:11,600 Speaker 5: and y coordinates for all players at ten or twenty 709 00:40:11,600 --> 00:40:13,840 Speaker 5: five frames per second, so that means I guess twenty 710 00:40:13,840 --> 00:40:18,600 Speaker 5: five rows for a single frame. We never really touch, 711 00:40:18,719 --> 00:40:21,440 Speaker 5: let's say the raw data in the sense that we 712 00:40:21,520 --> 00:40:23,560 Speaker 5: don't get the pictures of the game and then try 713 00:40:23,560 --> 00:40:25,759 Speaker 5: to figure out ourselves where the what the coordinates are 714 00:40:25,880 --> 00:40:28,319 Speaker 5: or what the coordinates are. So there are a lot 715 00:40:28,360 --> 00:40:31,160 Speaker 5: of data providers out there that either put cameras in 716 00:40:31,160 --> 00:40:35,719 Speaker 5: the stadium or use the broadcast footage to extract this data, 717 00:40:35,800 --> 00:40:38,160 Speaker 5: so they will have different models. One of the models 718 00:40:38,200 --> 00:40:41,799 Speaker 5: would be first identify where all the pitch markings are, 719 00:40:41,920 --> 00:40:44,920 Speaker 5: so they have a way to understand like where all 720 00:40:44,960 --> 00:40:47,560 Speaker 5: the players are relative to the pitch markings. Then they 721 00:40:47,600 --> 00:40:49,160 Speaker 5: know where all the players are and you can convert 722 00:40:49,200 --> 00:40:51,719 Speaker 5: all of that into coordinates. They know where the goals 723 00:40:51,719 --> 00:40:54,680 Speaker 5: are obviously, and then you'll get that in a single 724 00:40:54,719 --> 00:40:57,680 Speaker 5: file for a single game. Some providers might give you 725 00:40:58,440 --> 00:41:00,399 Speaker 5: one file per minute, they might give you one file 726 00:41:00,440 --> 00:41:03,400 Speaker 5: per half, and then that's just the raw data with 727 00:41:03,400 --> 00:41:05,360 Speaker 5: the identifiers in the cord that you might get an 728 00:41:05,400 --> 00:41:08,120 Speaker 5: additional file that has all the metadata that says this 729 00:41:08,280 --> 00:41:11,440 Speaker 5: identified blongs to this player, this identified blongs to this team. 730 00:41:11,560 --> 00:41:13,400 Speaker 5: If you're lucky and you're tracking data, you get an 731 00:41:13,400 --> 00:41:15,960 Speaker 5: identify that says which team is actually on the ball, 732 00:41:16,480 --> 00:41:20,359 Speaker 5: because that's highly relevant, but sometimes it's not included and 733 00:41:20,400 --> 00:41:22,720 Speaker 5: you have to figure it out yourself, like calculating conditions 734 00:41:22,760 --> 00:41:25,319 Speaker 5: to the ball for each player. And then obviously you 735 00:41:25,320 --> 00:41:28,600 Speaker 5: have skeletal data, which is I guess twenty seven times 736 00:41:28,680 --> 00:41:30,960 Speaker 5: more dense, or maybe even more, because you have twenty 737 00:41:31,000 --> 00:41:35,439 Speaker 5: seven coordinates, one for each body post point or body points, 738 00:41:35,440 --> 00:41:37,360 Speaker 5: so you might have one for your left shoulder, for 739 00:41:37,480 --> 00:41:38,960 Speaker 5: the tip of your nose, for your left ear, for 740 00:41:39,000 --> 00:41:41,280 Speaker 5: your right ear, for your right foot, for your ankle, 741 00:41:41,719 --> 00:41:43,839 Speaker 5: and that gets into the millions and millions of data points. 742 00:41:43,840 --> 00:41:46,120 Speaker 5: So when you talked about in the introduction about these 743 00:41:46,160 --> 00:41:49,560 Speaker 5: petabytes of data, I assume it's going to be mostly 744 00:41:49,560 --> 00:41:53,160 Speaker 5: skeletal data because that data is incredibly rich. 745 00:41:53,840 --> 00:41:57,080 Speaker 3: How do players actually feel about all of those because 746 00:41:57,200 --> 00:42:00,520 Speaker 3: you know, if people were watching me do my job 747 00:42:00,560 --> 00:42:05,560 Speaker 3: and monitoring like my next movie, they are all right, 748 00:42:05,640 --> 00:42:08,840 Speaker 3: but no one's modeling like what I'm doing with my 749 00:42:08,920 --> 00:42:12,280 Speaker 3: hands or my feet at all hours of this particular recording, 750 00:42:12,440 --> 00:42:14,920 Speaker 3: Like I would have mixed feelings about it, right, Like, 751 00:42:15,560 --> 00:42:18,760 Speaker 3: do you have any color on how players actually feel 752 00:42:18,760 --> 00:42:22,880 Speaker 3: about I guess the rise of a statistical analysis in football. 753 00:42:23,200 --> 00:42:27,000 Speaker 6: I think they find it useful. I think something that 754 00:42:27,800 --> 00:42:31,719 Speaker 6: yours has worked on this specifically is is eyesight. What 755 00:42:31,719 --> 00:42:34,120 Speaker 6: what can they see and what they cannot see? And 756 00:42:34,200 --> 00:42:37,520 Speaker 6: so when you look at the data and you think 757 00:42:37,520 --> 00:42:43,760 Speaker 6: about opportunity costs decision making, and you make a suboptimal decision, 758 00:42:44,480 --> 00:42:46,680 Speaker 6: when you go and talk to the player, they might 759 00:42:46,719 --> 00:42:48,719 Speaker 6: just simply tell you I couldn't see it, and then 760 00:42:48,760 --> 00:42:50,560 Speaker 6: you move on to you move on to the next. 761 00:42:51,160 --> 00:42:54,719 Speaker 6: So it's it's very complex. I find most players to 762 00:42:55,040 --> 00:43:00,200 Speaker 6: embrace the data. There's curiosity around it, but you know, 763 00:43:00,239 --> 00:43:02,200 Speaker 6: they likewise know the limitations. 764 00:43:02,360 --> 00:43:04,920 Speaker 2: You know, we started to talk about o soccer's fluid. 765 00:43:05,000 --> 00:43:07,439 Speaker 2: It's a beautiful game, it's an art. I totally agree, 766 00:43:07,600 --> 00:43:10,919 Speaker 2: but like people actually did say this about chess thirty 767 00:43:11,040 --> 00:43:13,240 Speaker 2: or forty years ago, and there was actually some people 768 00:43:13,239 --> 00:43:15,600 Speaker 2: who held out the belief computers will never be able 769 00:43:15,600 --> 00:43:17,879 Speaker 2: to beat humans at chess because of it as an art. 770 00:43:17,880 --> 00:43:20,680 Speaker 2: And it almost seems hilarious that like this view held 771 00:43:20,719 --> 00:43:23,160 Speaker 2: on for as long as it did. But it turns 772 00:43:23,200 --> 00:43:26,239 Speaker 2: out that no, chess is just a calculation problem, and 773 00:43:26,280 --> 00:43:29,520 Speaker 2: when you have enough data to compute, you could solve 774 00:43:29,600 --> 00:43:33,480 Speaker 2: the game. Kind of is soccer in the end? Like, 775 00:43:34,120 --> 00:43:37,040 Speaker 2: is it just a series of lots and lots of 776 00:43:37,600 --> 00:43:41,000 Speaker 2: discrete events that our eyes are not capable of, But 777 00:43:41,080 --> 00:43:44,319 Speaker 2: at the end of the day, with sufficient compute and 778 00:43:44,520 --> 00:43:48,000 Speaker 2: data collection, is it just like chess? And that it's 779 00:43:48,160 --> 00:43:52,200 Speaker 2: just a lot of micro binary decisions that can all 780 00:43:52,239 --> 00:43:54,799 Speaker 2: be summed up. And is this therefore how life is? 781 00:43:54,960 --> 00:43:57,480 Speaker 5: When you work with this data as long as I have, 782 00:43:57,600 --> 00:44:00,440 Speaker 5: I mean with this tracking data specifically, in the beginning, 783 00:44:00,480 --> 00:44:04,240 Speaker 5: it seems very overwhelming because it's just an infinite stream 784 00:44:04,360 --> 00:44:07,359 Speaker 5: of coordinates, and so I strove at that a little bit. 785 00:44:07,400 --> 00:44:09,560 Speaker 5: In the beginning, I was wondering what I should do 786 00:44:09,600 --> 00:44:11,799 Speaker 5: with this? How are you going to model any of this? 787 00:44:12,320 --> 00:44:14,520 Speaker 5: But I totally agree with you. You can discrictize this 788 00:44:14,640 --> 00:44:19,080 Speaker 5: so you can discertise it into let's say, very minor 789 00:44:19,239 --> 00:44:21,839 Speaker 5: events where we use the on ball event data, and 790 00:44:21,880 --> 00:44:23,840 Speaker 5: if we align that with the positional tracking data, we 791 00:44:23,920 --> 00:44:26,480 Speaker 5: might know every single moment where a player makes you pass, 792 00:44:27,080 --> 00:44:29,080 Speaker 5: and then you might know the next moment where a 793 00:44:29,080 --> 00:44:31,640 Speaker 5: player makes a reception, so that could be discretized into 794 00:44:31,640 --> 00:44:33,600 Speaker 5: one event that might last two and a half seconds 795 00:44:33,680 --> 00:44:36,759 Speaker 5: or three seconds. The next step would then be the 796 00:44:36,760 --> 00:44:39,960 Speaker 5: player receives the ball, they make an onball action that 797 00:44:40,080 --> 00:44:42,279 Speaker 5: also lasts two and a half seconds. That would then 798 00:44:42,360 --> 00:44:45,279 Speaker 5: be your next discratized event. And then you have all 799 00:44:45,280 --> 00:44:49,120 Speaker 5: these small events which are micro movements by players counter 800 00:44:49,200 --> 00:44:52,520 Speaker 5: movements by defenders. And if you go these let's say 801 00:44:52,560 --> 00:44:57,160 Speaker 5: sequences one at a time, you can still make aggregated 802 00:44:57,800 --> 00:45:01,040 Speaker 5: metrics from this using all the tracking data that you 803 00:45:01,080 --> 00:45:04,400 Speaker 5: have at your disposal, but it's actually still understandable for you. 804 00:45:04,560 --> 00:45:06,759 Speaker 5: So you might say, well, this player made this many 805 00:45:06,840 --> 00:45:10,880 Speaker 5: dribbles and it gained the team this much in terms 806 00:45:10,960 --> 00:45:13,680 Speaker 5: of added value to scoring a goal. And you can 807 00:45:13,719 --> 00:45:15,879 Speaker 5: do the reverse for defenders, where you can say, well, 808 00:45:15,920 --> 00:45:18,120 Speaker 5: this defender was always close to the ball, so he 809 00:45:18,280 --> 00:45:22,640 Speaker 5: was helping not concede a goal. And if you go 810 00:45:22,719 --> 00:45:24,960 Speaker 5: back to the analysis part where you want your video 811 00:45:25,000 --> 00:45:29,279 Speaker 5: analysts to look at this. They also do this just discriptized. 812 00:45:29,960 --> 00:45:32,959 Speaker 5: In discriptizing it like this will help significantly. 813 00:45:33,280 --> 00:45:35,480 Speaker 3: So I have one more question, which is it is 814 00:45:35,560 --> 00:45:39,160 Speaker 3: obviously World Cup season, which means it is also a 815 00:45:39,160 --> 00:45:45,040 Speaker 3: cell side analysts publishing World Cup prediction notes and research season. 816 00:45:45,800 --> 00:45:49,000 Speaker 3: And in my experience, they tend not to be very good. 817 00:45:49,320 --> 00:45:52,760 Speaker 3: Like often they will publish that England or the USA 818 00:45:53,000 --> 00:45:54,960 Speaker 3: are going to win whatever World Cup and. 819 00:45:54,960 --> 00:45:57,080 Speaker 2: As the numerista added, just predictor. 820 00:45:56,760 --> 00:45:59,840 Speaker 3: Pan that not to my knowledge, but you know, they 821 00:46:00,080 --> 00:46:05,480 Speaker 3: often get it wrong. If we think about statistical modeling data, 822 00:46:05,680 --> 00:46:08,280 Speaker 3: I mean, the cell side firms they should be pretty 823 00:46:08,280 --> 00:46:10,520 Speaker 3: good at this, and yet do you have a take 824 00:46:10,560 --> 00:46:14,120 Speaker 3: on why they seem to struggle with soccer predictions every 825 00:46:14,120 --> 00:46:14,840 Speaker 3: four years? 826 00:46:15,040 --> 00:46:18,400 Speaker 6: I like, honestly, for them, I have no idea what 827 00:46:18,840 --> 00:46:22,920 Speaker 6: data they're using, whether they're using an Elo model, whether 828 00:46:23,560 --> 00:46:28,040 Speaker 6: Nomura has on ball event data tracking data. There's different 829 00:46:28,080 --> 00:46:31,319 Speaker 6: ways that you can come up with these predictions. But 830 00:46:31,760 --> 00:46:33,640 Speaker 6: I think, you know, I like to stay stay in 831 00:46:33,680 --> 00:46:37,520 Speaker 6: your lane, and I think cell side analysts should stick 832 00:46:37,560 --> 00:46:39,000 Speaker 6: to cell side analyzing. 833 00:46:40,719 --> 00:46:43,400 Speaker 5: Simply. The age Old's problem. If you have a good model, 834 00:46:43,400 --> 00:46:46,560 Speaker 5: you wouldn't publish it. You would just beat the bookies, right. 835 00:46:47,000 --> 00:46:50,480 Speaker 2: Yeah, there's often criticism of people who publish things for 836 00:46:50,520 --> 00:46:52,920 Speaker 2: a living. Mike and yours, thank you so much for 837 00:46:53,040 --> 00:46:55,279 Speaker 2: coming on odd Laws. Learned a lot there and I 838 00:46:55,320 --> 00:46:57,200 Speaker 2: really appreciate your time and enjoy the rest of the 839 00:46:57,239 --> 00:46:57,640 Speaker 2: World Cup. 840 00:46:57,840 --> 00:47:11,520 Speaker 4: Thank you, Thank you, Tracy. 841 00:47:11,560 --> 00:47:13,840 Speaker 2: I have to admit I find it a little depressing 842 00:47:13,880 --> 00:47:17,640 Speaker 2: that probably most things in life are probably just computation. 843 00:47:18,040 --> 00:47:18,279 Speaker 6: Thanks. 844 00:47:18,360 --> 00:47:19,040 Speaker 1: You know what I'm saying. 845 00:47:19,080 --> 00:47:21,279 Speaker 2: It's like a lot of thing that like that there 846 00:47:21,320 --> 00:47:24,200 Speaker 2: is something called art and beauty and intuition and something. 847 00:47:24,680 --> 00:47:27,720 Speaker 2: It's probably just computers all the way down. Binary events 848 00:47:27,760 --> 00:47:30,600 Speaker 2: that can be chunked and analyzed by microchips. 849 00:47:30,600 --> 00:47:32,920 Speaker 3: Oh you know the question I would have asked, Yeah, 850 00:47:32,920 --> 00:47:36,680 Speaker 3: but we ran out of time was the idea of 851 00:47:36,719 --> 00:47:39,440 Speaker 3: like good Art's law, which is like once you have 852 00:47:39,520 --> 00:47:43,200 Speaker 3: a measure, like the measure. Yeah, because you see this 853 00:47:43,320 --> 00:47:47,239 Speaker 3: criticism of sports analytics, coaches like players start focusing on 854 00:47:47,280 --> 00:47:50,440 Speaker 3: their stats. Coaches are focusing on the player stats, and 855 00:47:50,480 --> 00:47:52,520 Speaker 3: then they get the players with the good stats, and 856 00:47:52,560 --> 00:47:54,920 Speaker 3: then they just focus on improving their stats more. But 857 00:47:54,960 --> 00:47:58,480 Speaker 3: the stats don't necessarily translate into like wins all the time. 858 00:47:58,680 --> 00:47:59,319 Speaker 1: Yeah, you know. 859 00:47:59,400 --> 00:48:01,960 Speaker 2: I was thinking something that Mike said in the beginning, 860 00:48:02,000 --> 00:48:04,680 Speaker 2: which is that like a particularly like an English Premier 861 00:48:04,760 --> 00:48:07,839 Speaker 2: League team, there's multiple things they could be optimizing for. 862 00:48:07,960 --> 00:48:10,480 Speaker 2: So they could be optimizing for profit, it could be 863 00:48:10,520 --> 00:48:14,840 Speaker 2: optimizing for avoiding relegation, and they could be optimizing for 864 00:48:14,960 --> 00:48:18,600 Speaker 2: avoiding relegation. They could be optimizing for wins. Those are 865 00:48:18,719 --> 00:48:22,120 Speaker 2: all distinct things. But you see this when a sport 866 00:48:22,160 --> 00:48:25,160 Speaker 2: gets over optimized. It didn't come up with a good example. 867 00:48:25,280 --> 00:48:27,480 Speaker 2: Is like basketball. When I was younger, it was like 868 00:48:28,120 --> 00:48:30,440 Speaker 2: the game was fun because there were lots of slam dunks, 869 00:48:30,600 --> 00:48:33,319 Speaker 2: and then everyone realized that three point attempts were not 870 00:48:33,360 --> 00:48:36,400 Speaker 2: being taken enough. So suddenly the game is dominated by 871 00:48:36,440 --> 00:48:38,800 Speaker 2: three pointers, which may be a better way to play, 872 00:48:39,000 --> 00:48:42,719 Speaker 2: but it's not necessarily a more fun fan experience than 873 00:48:42,760 --> 00:48:45,239 Speaker 2: watching a dunk. So you think, like, okay, could the 874 00:48:45,280 --> 00:48:48,560 Speaker 2: game get better formally, but it becomes less entertaining. There's 875 00:48:48,560 --> 00:48:49,280 Speaker 2: a lot of criticism. 876 00:48:49,280 --> 00:48:51,879 Speaker 3: I think that's the possibility because people are already talking 877 00:48:51,960 --> 00:48:55,880 Speaker 3: about convergence and like actual play style. 878 00:48:55,800 --> 00:48:58,000 Speaker 2: Totally, and you see this, like you know, there was 879 00:48:58,040 --> 00:49:01,440 Speaker 2: a lot of criticism in like, people are very critical 880 00:49:01,480 --> 00:49:04,480 Speaker 2: of how Paraguay played right, like play to just survive 881 00:49:04,560 --> 00:49:06,920 Speaker 2: into penalty kicks, then hope that the variance of the 882 00:49:06,920 --> 00:49:09,920 Speaker 2: penalty gap period allows you to beat France. But it's like, no, 883 00:49:10,040 --> 00:49:12,759 Speaker 2: that's like the game theory optimal or just the game 884 00:49:12,800 --> 00:49:16,080 Speaker 2: optimal play if you're considered to be the weaker team. 885 00:49:16,200 --> 00:49:20,000 Speaker 2: So it does feel like there's all different things, you know. Again, 886 00:49:20,080 --> 00:49:22,719 Speaker 2: going just to the core of the question, stat's like, 887 00:49:22,960 --> 00:49:26,040 Speaker 2: what are you solving for? Solving for winning is very 888 00:49:26,080 --> 00:49:28,480 Speaker 2: different from solving for profit, is very different from solving 889 00:49:28,520 --> 00:49:32,160 Speaker 2: for a gambler, And there are different answers to each one. 890 00:49:32,239 --> 00:49:34,360 Speaker 3: Yeah, you win, but no one's paying for it or 891 00:49:34,360 --> 00:49:35,200 Speaker 3: happy about it. 892 00:49:35,840 --> 00:49:38,799 Speaker 2: Plausible, although for now people are paying crazy amounts of 893 00:49:38,800 --> 00:49:41,040 Speaker 2: money still to go see a soccer games. 894 00:49:41,120 --> 00:49:42,480 Speaker 3: So all right, shall we leave it there. 895 00:49:42,520 --> 00:49:43,200 Speaker 5: Let's leave it there. 896 00:49:43,360 --> 00:49:45,600 Speaker 3: This has been another episode of the All Thoughts podcast. 897 00:49:45,680 --> 00:49:48,840 Speaker 3: I'm Tracy Alloway. You can follow me at Tracy Alloway. 898 00:49:48,440 --> 00:49:51,400 Speaker 2: And I'm Joe Wisenthal. You could follow me at the Stalwart. 899 00:49:51,520 --> 00:49:54,560 Speaker 2: Follow our producers Carmen and Rodriguez at Carman armand dash 900 00:49:54,560 --> 00:49:58,120 Speaker 2: Ol Bennett at Dashbock, Kilbrooks at Kilbrooks and Kevin Lozano 901 00:49:58,160 --> 00:50:01,120 Speaker 2: at Kevin Lloyd Lozano. From our Odd Lots content, go 902 00:50:01,160 --> 00:50:03,520 Speaker 2: to Bloomberg dot com slash odd Lots or the daily 903 00:50:03,560 --> 00:50:05,919 Speaker 2: newsletter and all of our episodes and you can shut 904 00:50:05,920 --> 00:50:07,759 Speaker 2: about all of these topics twenty four to seven in 905 00:50:07,920 --> 00:50:11,000 Speaker 2: our discord Discord dot gg slash odlines. 906 00:50:11,080 --> 00:50:12,960 Speaker 3: And if you enjoy Oddlots, if you like it when 907 00:50:12,960 --> 00:50:16,279 Speaker 3: we talk about stochastic soccer modeling, then please leave us 908 00:50:16,320 --> 00:50:19,720 Speaker 3: a positive review on your favorite podcast platform. And remember, 909 00:50:19,760 --> 00:50:22,080 Speaker 3: if you are a Bloomberg subscriber, you can listen to 910 00:50:22,200 --> 00:50:25,120 Speaker 3: all of our episodes absolutely ad free. All you need 911 00:50:25,160 --> 00:50:27,520 Speaker 3: to do is find the Bloomberg channel on Apple Podcasts 912 00:50:27,560 --> 00:50:30,240 Speaker 3: and follow the instructions there. Thanks for listening 913 00:51:00,040 --> 00:51:01,000 Speaker 4: And