Today I am going to try a bit of a different kind of post. I am fond of saying that we’re trying to build a community here, and I think this piece will make that hopefully a little closer to reality. Today I’m going to talk about some upcoming research and pieces I am working on, and invite people to share if they have anything they think could improve on or just any ideas in general. I have gotten great feedback in the past, and I figured making that call a little more explicit today might help us go even farther on our sumo analytics journey. So I’ll share the topics bolded, and then explain a little, and if you feel there’s something missing or to be bettered, then please reach out on various channels be it twitter, the blog itself, email, or even on the reddit. Without further ado and not particularly ordered:
Ranking the Best Sumo Runs of All Time
MJ vs Lebron, Brady vs Manning, Ali vs Lewis; one of the most popular sports discussion topics is always: who is and was the best? We’re a bit robbed of that in sumo as Hakuho’s credentials are just about unassailable. Even still, how does Haramafuji’s best run stack up next to Musashimaru’s? That’s a question a bit tougher to answer, and one I am interested in.
It’s probably no surprise given the content of this blog, but I’m a fan of baseball and of Bill James’. He’s the man that in many ways pioneered analytics for baseball, and in that sense is the godfather to all analytics. He has a baseball dynasty ranking system that I’m inserting below, and then trying to convert into sumo. Please note that for this there are 162 games in the baseball season in case you’re not familiar.
6 points: Win 100 games AND the World Series
5 points: Win the World Series
4 points: Win 100 games AND the pennant
3 points: Win 100 games AND the division OR win the pennant
2 points: Win 100 games OR the division
1 point: Win 90 games OR make the playoffs
-2 points: Fail to make the playoffs
-3 points: Losing record
Any team that hits 10 points (or more) can be considered a dynasty. 0 points resets a run.
A quick point: this isn’t some gold standard so to speak. He chose some nice round numbers (100 wins or 90 wins) and some arbitrary accomplishments (why is winning the division and 100 games equal to the pennant for 3 points?). As a math/stats/finance note, when we use arbitrary it isn’t necessarily negative but rather essentially means anything. It’s arbitrary that winning 100 games and the World Series is 1 point more than winning the World Series alone, but that doesn’t mean it’s bad or wrong. It just means it was a decision made that could have been made otherwise. So my upcoming proposed standards are arbitrary (as is the exercise), but I hope that it lines up well with what we think makes an impressive run of sumo (and is fun!) which is why I’m opening up for feedback.
My initial thinking is to simply update the above points with relevant tournament accomplishments and then simply divide by 6 (there are 6 tournaments a year, vs 1 season of baseball a year) but honestly I can just keep the points and instead multiply the dynasty threshold by 6, so 60 points for now. We might adjust that later once I have the rankings, but that’s our starting point. How would one get those 60+ points in the first place? Here’s my current criteria I’ve spent a moderate amount of time contemplating and 0 time actually applying to sumo (yet).
6 points: Zensho Yusho (15-0 Yusho)
5 points: Yusho
4 points: Jun-Yusho
3 points: 12+ wins
2 points: 10+ wins
1 point: Kachikoshi (8+ wins)
-3 points: Makekoshi (8+ losses)
I think also make it additive as in if you got a Yusho + 13 wins that’d be 5 + 3 points. Again, these points can be essentially anything but I thought it’d be nice to open it up to the community and think about this some beforehand rather than letting our first batch of results influence us after the fact.
The only other thing I was thinking was, potentially giving wrestlers points for beating Yokozuna or guys that were future Yokozuna (or perhaps Ozeki). I think it’s good because it rewards wrestlers for having impressive wins. On the other hand, if you do it with current Yokozuna, you’re saying that beating Ozeki Hakuho in May 2007 is worth less than beating him 2 months later in July when he was Yokozuna. If you do based on future Yokozuna, then that means it could be a decade+ before you know how impressive a tournament win was (in this ranking system).
As an additional note, I do think that if this doesn’t end up with Hakuho as the leader of this, then we did something wrong. I’m joking a little, but also not really.
Anyways, please reach out if you think there’s a better way to do this or if you have any ideas for improvements here.
We Know of the Joi, What About Other Sumo Levels?
My current larger research project is predicting daily matches in the Makuuchi - i.e. on day 1, who will the Yokozuna face, who will Ozeki 1E face, etc. At the top of the Banzuke, we have a term for who might reasonably expect to face the Yokozuna each tournament and that is the Joi. However, my next goal would be to try and define those sorts of bands or ranges or whatever you prefer to call them further down the Banzuke. Like the second “grouping” would be potentially from Maegashira 3-10 then Maegashira 11-17 and Juryo 1-7 and so on.
This would be helpful because then we can start indexing wrestlers against this. For instance we know a wrestler’s prime is about 27 when you’re in the Juryo and Makuuchi, so I’d like to try and define “true" talent levels. One way I was thinking of was to look at how often in any given “grouping” a wrestler is between 6-9 to 9-6 (not dominant, but not overmatched) and use that to find people whose primes are mid-Maegashira, or upper-Juryo and so on.
This would also be helpful for development questions too. One thing I’ve noticed is that a lot of wrestlers hit a wall in the upper Makushita, say Makushita 20 or so and better. But this would give us a better defined language to test where are common developmental hurdles for wrestlers.
Those were just a few ideas I had, but if you have other ideas on how to define true talent level, or to explore those developmental questions, I’d love to hear!
Who Will Win Individual Matches?
This one is obviously a massive can of worms. There’s actually a lot of different ways I want to look at this in no particular order:
If 2 wrestlers fight each other, at a certain point does a wrestler winning more become predictive of future matches between them? (i.e. if Ura beats Abi 5 times in a row, do we expect him to win 6 times in a row and so on?)
Do certain wrestling styles do well against certain others?
Are wrestlers “streaky” - if you won your last 3 matches does that make you meaningfully more likely to win your next, for instance
I can’t find the paper, but I am aware that there was a 1985 paper “debunking” the hot hand in basketball. However, there has been more research showing why the ‘85 paper may have been flawed, so we would be on semi firm ground here at least
Going back to the Elo work
I honestly have the least idea on how I want to tackle this problem over any that I intend to explore, so if you have anything please please reach out on this one haha.
Sumo Stories
This one would be building off of my posts where I go a little more in-depth into wrestlers; an example being Hoshoryu here1. However, I think there’s a lot of room to tell stories beyond something like that. A few ideas I have are the greatest Ozeki tournament ever (as in the Ozeki all did well, and several were future Yokozuna - I don’t want to give it away yet but you could probably find it and if you message me I’ll confirm), the mini wave of Russian wrestlers in the late 2000’s and their subsequent downfall, honing in on an old tournament and telling the story of some key players - ideally one where I can have footage too like this terrific Akebono Yusho. If you have any ideas, feel free to hit me up with them! That said, these might be for later as I would ideally like to be a little better at Japanese and be able to value add by digging into old articles.2
Office Hours and/or Ask Me Anything
This one is only if there’s actual demand. But if there was, I think it could be fun to just answer questions. That said, I also am not a genius or anything so who knows if anyone is interested - that’s why I’m asking! I also might eventually look to see if any of the podcasts would have me (I’m a big fan of Grand Sumo Breakdown myself but like all of them), but I also probably don’t have enough material to be an interesting guest so I’ll keep you posted if I do ever make it outside the confines of this comfy community.
Conclusion
These are some ideas I had and at least the start of how to attack them, but if you have further research questions, then feel free to reach out.
Thanks for reading! Hopefully this was interesting or helpful for me to pull the curtain back a bit.
As a final note, going forwards, I’ll be trying to post every Monday morning. I was already weekly so the only change here will be that you can just expect the posts on Mondays.
I really should do my weekly post so I’ll try to do some of these going forwards when I don’t have a longer weekly post ready
My Japanese is so-so reading but I’m working at it a little harder lately and keep up regardless
Aloha, Michael,
I tried writing to you in a couple of different ways, but I'm not sure my notes made it through.
Thank you for your posts. I really enjoy reading them.
You wrote:
> I can’t find the paper, but I am aware that there was a 1985 paper “debunking” the hot hand in basketball. However, there has been more research showing why the ‘85 paper may have been flawed, so we would be on semi firm ground here at least
Check out Leonard Mlodinow's book The Drunkard's Walk. Mlodinow co-authored books with Stephen Hawking. This book has a number of examples debunking winning and losing streaks in everything from gambling to baseball to movie executives. It's a very interesting and informing book.
Keep up the good work.
Cheers,
John
Some notes:
Your dynasty system hands out points too freely and deducts them too rarely(even assuming that you are counting kyujos as MKs which you should). It devolves into a lifetime ranking system, as basically no Yokozuna is going to fall below 0 once they rise above it. Hakuho and Musashimaru, just as a couple examples off the top of my head, went 8+ years competing in every tournament with no MKs. It's probably just more useful to define individual categories(3 tournament peak, 1 year peak, 3 year peak, 5 year peak, 10 year peak) and find the best such runs. I wouldn't be surprised if Hakuho leads all of them, assuming you are only counting modern era.
Creating the elo system is going to be the hardest part, but from there calculating streaks should be quite straightforward. Just take a look at a bunch of H2H matchups and see whether there's a trend where rikishi with better H2Hs outperform their elo.
For best Ozeki basho I assume that you mean either 1963.07 or 2012.07 depending on what you value.(psychologically speaking, I assume you mean 1963.07 as 2012.07 wasn't that long ago and pretty famous)
I'm also going to add some pretty well-known stuff about extra Sekiwake slots that your recent article gets wrong:
Since about 2006 or so, the banzuke committee got much stricter about handing out extra Sekiwake slots. The only real way to do it, is to get 11+ wins from Komusubi. Some counterexamples for other rank/records:
10win Komusubi: https://sumodb.sumogames.de/Query.aspx?show_form=0&form1_rank=K&form1_wins=10&form1_year=2006-2024
4 10-win Komusubis fail to get promoted, with none opening up a new spot.
12+ high Maegashira: https://sumodb.sumogames.de/Query.aspx?show_form=0&form1_rank=M1-M3&form1_wins=12-15&form1_year=2006-2024
Three out of the last six to do so landed at Komusubi, including Daieisho after 10-5 from M2 and 13-2 from M1, denying him a chance at an ozeki run. The only counterexample here is Takayasu. The general consensus is that with Shodai being demoted that basho, there were only two Y/O left: Terunofuji and Takakeisho, both struggling with injury. Takayasu had 23 wins over his last two bashos and they wanted to give him a chance to win back Ozeki and hopefully bring some stability.
Two other one-off occurences, both involving Covid. When Asanoyama lost his Ozeki rank during his demotion, he would have normally taken the second Ozeki spot. However, the banzuke committee felt that it would have been unfair to deny someone a Sekiwake promotion due to Asanoyama's punishment, as normally he would have retained his Ozeki spot and a spot would have been open, so Meisei was promoted with a 8-7 from K as if Asanoyama was not there.
Similarly, Daieisho's Sekiwake spot was frozen when he was forced to pull out due to Covid with a 6-6. Since he didn't actually keep his Sekiwake spot through wins, they treated it as opening up a new spot to let him keep his rank, and Hoshoryu was promoted with a 9-6.
In the 1990s they gave out extra Sekiwake spots like candy, but times have changed and 11+ wins from Komusubi is the only reliable way to open a new Sekiwake spot