Rank Changes Through The Years

While we put together our own predictions of the next banzuke, inevitably we’ll come across some real head scratchers. Tim gave me some insight on Twitter that kyujo complicates this banzuke guessing-game by robbing us of our zero-sum game. There will be absences, like those of Terunofuji, which will not have corresponding wins among the rest of the field. So, inevitably, we get some wild swings especially in the mid-maegashira ranks. I’ve attempted to visualize those swings.

I’ve been toying around with a way to visualize this in a way that will be helpful for trying to predict the banzuke. So what I’ve got right now is a little dynamic heatmap that takes each wrestler’s starting rank (going back to 1958 and the modern 6 basho, 15 match-day calendar), and plots it along side where they end up on the next banzuke.

I don’t have 2022 data in here yet and that’s not really a priority at the moment. In fact, you may want to filter the years to exclude the “Covid Years” because this shows the effect of Covid Kyujo. The histogram of wins shows that there’s a bit of skew, a fat tail there, for wrestlers with zero wins as we see that instances of full kyujo occur more often than wrestlers getting one or two wins in a tournament. Click on that 0 win bar, and the outliers are apparent as some wrestlers preserved their rank with zero wins, while most dropped dramatically. Around M5, they’re all dropping to Juryo but we can see that even wrestlers in the joi have fallen out of the top division by posting zero wins.

Zero Wins: Narrow Band of Possibility, Unless Covid Kyujo

“Well, what does this mean for my predictions, Andy?” There are several things we can learn. For example, if a wrestler has 6 or 7 wins, or even 11 or 12 wins, there’s a very narrow band of probable ranks to put that wrestler. However, if a wrestler has 8 or 9 wins, this appears to be where the art comes into play. With 8 wins, a wrestler at Maegashira 11 has been promoted anywhere from Maegashira 2…to staying on that same 11th rank. The heatmap shows us they tend to fall around M8 but specific circumstances can make for wild swings.

Wider Bands for 8-win Kachi-Koshi

We’d expect a relatively normal distribution of wins but there’s actually fewer instances of 7-wins than expected, even fewer than there are 6-win records. There’s also a bit of a larger-than-expected spike for 8 wins. When filtered for 2012-2021, the distribution appears more normal. There’s still a spike at 8 wins but with the importance of that 8-win threshold on a sekitori’s career and income, it is to be expected, especially if one is possibly injured; Get that 8th win and then go kyujo. But is this evidence that yaocho has been rooted out of the sport, or at least that the situation is better than it was decades ago?

Less Pronounced 8-Win Spike

My Guess

Now, how did this help (or not help) my banzuke guess? Well, we won’t really know until the banzuke comes out but I feel like I did my best. I wasn’t as leery of over-promoting wrestlers with 8 and 9 wins, of which there were several. It was still very difficult to place Takayasu. From his position he could very well contend for a sanyaku position but I had to settle him at Maegashira 2 East. This would be a historically low position for M7 with 12 wins. He was in this same position after Hatsu 2013 and was promoted to M1. Unfortunately, I think Ichinojo and Kiribayama will just pip him to those top rank-and-file slots since Takayasu was out of the joi-jin.

I think this did help me feel more confident in how I placed the Juryo promotees, and who I picked, as well. But there are definitely several who I am really not sure about. I had Shimanoumi as high as M5-6 for a while but I am more comfortable with this version. Tobizaru gets a decent bump but he’s certainly not far out of the ordinary.

I have published the visualization below here, just click the “Read More” link if you’re on the homepage. If you clicked through directly to the article from Twitter, it should show up by default, I believe. Anyway, I don’t want it to always load for everyone because I know the Tableau vizzes can churn for a bit. So for those just browsing on the home page, this won’t kill anything.

Anyway, if you click through and play with it, I would advise clicking the button to view it full screen, down toward the bottom right of the visualization. It makes it much easier to work with.

I’m also planning some updates. In particular, I want to include a parameter for the number of kyujo bouts in makuuchi. And I’ve got the data to do this for the lower divisions.

The Dashboard

15 thoughts on “Rank Changes Through The Years

  1. Very interesting stuff and I hope your prediction is close to the real basho. Is there a way of filtering out the wrestlers who made use of the “kosho” rule: if a man was injured during the basho he could sit out the next one with rank protection (I think thats right). Given the hugeness of the data set it probably wouldn’t make much difference, I suppose.

    • Yes! It does! Brilliant. I’d forgotten that. It would be the same method, just from years before COVID. So my image includes those, as well, since the years went back far enough. If I trim the years to pre-2019, we see the kosho rank preservation by clicking on the zero wins. The data at the bottom gives the specifics if you click on the heat map.

  2. This is awesome. As a veteran banzuke forecaster, I’ll venture that there are a couple of issues that make using the info for prediction tricky. First, and most obviously, rikishi compete for banzuke positions, so while it may be plausible for Tochi to be ranked M8 and Shimanoumi M9 in isolation, there’s no way that M10e 8-7 will be ranked below M15w 9-6. Second, the banzuke committee changes how it makes decisions over time based on its composition and possibly other factors we don’t know about. So while it used to be possible for well-performing juryo rikishi to be ranked as high as you have them, I’ll be shocked if that ends up being the case this time, based on where they have been landing recently. I’ll post my own prediction, which will differ from yours by several ranks for quite a few rikishi, but those are a couple of things that jumped out at me in using an approach that tries to place everyone at the position that rank-record combination is most likely to have landed them historically.

    • Thank you! I’m glad you like it. It sure does illustrate just how wild banzuke luck can get. There’s a lot more to it so I hope to add to it over time.

    • Re historicity, I would go as far as to say that the reason a “spread”, or distribution, of future banzuke position exists for a certain rank-record combination exists in the first place, is the need to adjust for the Law of Banzuke Physics: No two rikishi can occupy the exact same rank at the same time. The spread isn’t a reflection of where you’ll end up with a certain rank-record combination, it’s a reflection of the margin of error when you have another rikishi who ought to share the same rank as you do.

      Compare yokozuna and ozeki positioning, which is extremely clear cut and predictable because the ranks, while differing in prestige, are functionally equal in all other respects.

  3. Hello all, long time reader, first time poster here.
    I love that Sumo fans do things like this, the data you never knew you wanted until it was compiled and presented ;)
    Last Basho was the first time I tried to predict the Banzuke and came up with something pretty close to what you had. Totaling the rank discrepancies of each Rikishi put my error at 14 full ranks as compared to the official Banzuke. This time around though wow what a difference, the only thing we agree on below Hokutofuji on exact placement are Yutakayama and Midorifuji (excited to see his return, hope we’re both right on that). In total our predictions differ by over 46 full ranks. I went lenient on big make-koshi demotions like played out last time around, even Juryo’ing Kotokuzan and Kagayaki .vs. the Chiyo pair. Let’s see if the Banzuke committee is as deviously conservative as me >=).

    Terunofuji Y
    Mitakeumi O Shodai
    O Takakeisho
    Wakatakakage S Abi
    Hoshoryu K Daieisho
    Ichinojo M1 Kiribayama
    Takayasu M2 Kotonowaka
    Tamawashi M3 Endo
    Hokutofuji M4 Takanosho
    Onosho M5 Tobizaru
    Wakamotoharu M6 Ura
    Takarafuji M7 Shimanoumi
    Kotoeko M8 Terutsuyoshi
    Kotoshoho M9 Okinoumi
    Chiyoshoma M10 Sadanoumi
    Aoiyama M11 Myogiryu
    Tochinoshin M12 Nishikigi
    Meisei M13 Chiyotairyu
    Ishiura M14 Oho
    Azumaryu M15 Yutakayama
    Midorifuji M16 Ichiyamamoto
    Chiyomaru M17 Chiyonokuni

      • Yeah this is very close to mine, with some differences M10 and lower. Kotokuzan should be completely safe though; you don’t drop to juryo with a 7-8 from a whole rank higher than the bottom of the banzuke.

  4. … late on guess the banzuke, I am posting this here just because it is wild and… well… more often the not the banzuke committee is even wilder!
    I know I will regret this the moment I post it… too late in any case.

    Y Terunofuji

    O Mitakeumi Shodai

    S Wakatakakage Abi

    K Hoshoryu Daiesho

    M1 Ichinojo Kiribayama

    M2 Takayasu Kotonowaka

    M3 Endo Tamawashi

    M4 Hokutofuji Takanosho

    M5 Onosho Ura

    M6 Tobizaru wakamotoharu

    M7 Takarafuji Kotoeko

    M8 Shimanoumi Terutsuyoshi

    M9 kotoshoho Okinoumi

    M10 Tochinoshin Meisei

    M11 Aoiyama Myogiryu

    M12 Chiyoshoma Sadanoumi

    M13 Chiyotairyu nishikigi

    M14 Ishiura ichiyamamoto

    M15 Yutakayama oho

    M16 azumaryu midorifuji

    M17 kotokuzan Chiyomaru

  5. I’d caution against trying to extend this type of analysis too far into the lower divisions, because a major driver of the banzuke mechanics down there is the number of outright intai freeing up spots for everybody lower down to move up into. Starting from mid-sandanme or so, that’s likely to drown out all other effects.


This site uses Akismet to reduce spam. Learn how your comment data is processed.