Accumulated wins as an in-basho performance metric

[BroadMeadow]

Is there a way to gage a rikishi’s performance mid-basho? Certainly, there a numerous ways- but are there measurements that could expose “market inefficiencies” hidden by the simple won-loss record (W-L) of the wrestler? I propose testing the in-basho performance by the measurement of accumulated wins (AW).  Every match in which a rikishi defeats an opponent with any victories, those victories are accumulated to the winning rikishi. For example: at the end of day 1, there will be no accumulated wins because all of the victors defeated opponents with no wins. At the end of day 2, there will (almost certainly) be some wrestles with one accumulated win because some wrestlers will defeat wrestlers who won on day one. At the end of the second day, it is possible for a rikishi to have 3 AW if he defeated a 1-0 wrestler on day 2 and a 2-0 wrestler on day 3. If a wrestler faced and defeated an undefeated wrestler every day of the basho, he would end the tournament with 105 accumulated wins.

Granted, if 2 undefeated rikishi face off on day 15 it does not take a genius to surmise they are the best wrestlers in that tournament. At the other end of the scale, a wrestler with fewer than 3 wins entering day 15 has had a rough go of things. I anticipate this metric offering the most help in the “big middle” of the standings. This metric is also going to be more helpful as the basho progresses. (But again, as the basho progresses, we tend to know who is doing well and who is not- or we like to think we do.)

We are now just over half way through the Hatsu basho. What does the AW tally show us? Hakuho, the only undefeated rikishi, leads with 18 AW. In second place is Takakeisho (6-2) with 13 AW. Interestingly, Tamawashi is also 6-2, but only has 9 AW. Ahead of him are Kaisei (7-1 and 12AW); Yago (7-1 and 11); Abi (5-3 and 10); and Sadanoumi (4-4 and 10). Even though he has 2 fewer wins than Tamawashi, Sadanoumi has higher quality wins: he has beaten rikishi with better W-L records.

This is an in-process endeavor and maybe nothing profitable will come of it. To begin testing it, I will use it to look for discrepancies camouflaged by W-L record. Tomorrow’s schedule offers 1 match in which the AW opposes the W-L record of the rikishi. Myogiryu (3-5) and Ichinojo (5-3) have mirrored records, but by AW Myogiryu has been performing better this basho (9 v. 6) and should win. Using AW as a predictor, this is the only match in which the rikishi with the worse W-L record is expected to win.

The most evenly matched wrestlers tomorrow are Endo v. Onosho (8 each); and Takayasu v. Tamawashi (9 each). The greatest mismatches tomorrow are Hakuho v Kotoshogiku (18 v. 5); Takakeisho v. Nishikigi (13 v. 1); and Kaisei v. Ryuden (12 v. 0). Nishikigi is an interesting case which might help prove the usefulness of this metric. While his record is a respectable 4-4, he only has 1 AW. Contrast him with the aforementioned Sadonoumi who has the same record, but 10 AW.

[Yamanashi]

This reminds me of a "strength of schedule" metric used by (among others) NCAA football and basketball.

It also makes me think that the big winners in this metric will be those guys who go 5-10 but pick up a kinboshi.  I think Takamiyama used to specialize in that. Of course, if you got a kinboshi from Kisenosato last year, you weren't getting any AW's.*

 

* Sorry ... too soon?

Edited January 20, 2019 by Yamanashi
More thinking!

[itchyknee]

Accumulated losses would be important too.  Otherwise a 'lucky' victory over Hakuho would put an MK rikishi in the lead.

AW - AL, or should the equation be more complex than that?

[BroadMeadow]

10 minutes ago, itchyknee said:

Accumulated losses would be important too.  Otherwise a 'lucky' victory over Hakuho would put an MK rikishi in the lead.

AW - AL, or should the equation be more complex than that?

Thanks for that. I think a similar thought had crossed my mind some time ago when I was pondering these things!

[Tsuchinoninjin]

Yeah, its probably not a bad way to track this as a simple method. One nice thing is that it probably quickly de-rates rikishi who are fighting with a big injury. Of course that's usually easy to spot anyway but its nice to build it into your algorithm when you're entering thousands of games.

You might want to normalize this on head to head records though.

And, I guess your favorite permutation of ELO would probably be slightly better.

[itchyknee]

46 minutes ago, BroadMeadow said:

Tomorrow’s schedule offers 1 match in which the AW opposes the W-L record of the rikishi. Myogiryu (3-5) and Ichinojo (5-3) have mirrored records, but by AW Myogiryu has been performing better this basho (9 v. 6) and should win. Using AW as a predictor, this is the only match in which the rikishi with the worse W-L record is expected to win.

By AW-AL Ichinojo is predicted to win:

Under a forward looking AW AL scenario (where the wrestler gets the benefit and detriment of future performance of their opponent:

Myogiryu AW = 9, AL = 14 (not counting withdrawals and fusens)

Ichinojo AW = 12, AL = 4 (not counting withdrawals and fusens)

Edited January 20, 2019 by itchyknee
added alt text to image; forward looking AW/AL

[Nantonoyama]

Given that this topic's title is a great title for a scientific publication, if we formalise the theory enough in-depth, we may consider to write a paper and submit it to the International Journal of Sumo Gaming

[Dwale]

I am curious if this would work at all or not, but there is a certain random element to sumo that causes me to suspect that any such formula is going to be pretty fluky. 

[Gurowake]

My personal gut opinion is that this metric has very little predictive power beyond that which is inherent in the fact that rikishi who have already won matches tend to be the kinds of rikishi that will win matches in the future.  It's a distilling of available information into a single number, but I don't think it's done in a way that's particularly meaningful as it neglects a bunch of other information.

[Pandaazuma]

Well, I know one way to test this, @BroadMeadow

http://99998271.com/sb/

Click on 'intro'.

[yorikiried by fate]

10 hours ago, Nantonoyama said:

Given that this topic's title is a great title for a scientific publication, if we formalise the theory enough in-depth, we may consider to write a paper and submit it to the International Journal of Sumo Gaming

Knit in gender and the academic world is yours.

 

E.g.: Accumulated wins as an in-basho performance metric in safe spaces.

Edited January 21, 2019 by yorikiried by fate
example added

[BroadMeadow]

12 hours ago, Dwale said:

I am curious if this would work at all or not, but there is a certain random element to sumo that causes me to suspect that any such formula is going to be pretty fluky. 

I agree. On a match-to-match basis it seems so completely unpredictable, but taken as a whole there must be patterns and trends. We know Hakuho is going to win 12-15 matches and Goiedo is going to be either 8-7 or 7-8. Now if we can only figure out how!

[BroadMeadow]

9 hours ago, Pandaazuma said:

Well, I know one way to test this, @BroadMeadow

http://99998271.com/sb/

Click on 'intro'.

Its all fun and games and speculation...until there's money on it!

[BroadMeadow]

2 hours ago, yorikiried by fate said:

Knit in gender and the academic world is yours.

 

E.g.: Accumulated wins as an in-basho performance metric in safe spaces.

Very funny.

There is a limit on how many "reactions" you can have per day? Guess your smiley face will have to wait until tomorrow.

[Pandaazuma]

Just now, BroadMeadow said:

Its all fun and games and speculation...until there's money on it!

I dunno! I'd hate playing the games for money, although I'll have the occasional flutter on the gee-gees. I play the sumo games because it's not only a great and challenging tactical contest against some extremely competent people, but also an enjoyable way to learn more about sumo in general. Hope you'll give them a try...and test out your theory. ;

[yorikiried by fate]

6 minutes ago, BroadMeadow said:

Very funny.

There is a limit on how many "reactions" you can have per day? Guess your smiley face will have to wait until tomorrow.

Awarding someone a smiley face is somewhat condescending and therefore a micro aggression.

[BroadMeadow]

Thanks for all the input, I appreciate it. I think I need to spend some time with @itchyknee it looks like he is way ahead of me. But this sport (probably) has the longest historical statistical database of any now in existence, and I am so new to the game.

DAY 1 REPORT

Day 1 was a mixed bag. As some have commented and hinted at, this metric doesn't seem to advance much beyond simple W-L records. The over-all predictive record for AW on day nine's action was 9-7-3. The three are: the first match involving the juryo wrestler and two matches where rikishi had the same AW total. A 56% winning percentage is not bad.

Before the matches, I highlighted one bout where the AW record contradicted the W-L record: Ichinojo v. Myogiryu. The AW indicated Myogiryu would win, even though his record was worse than Ichinojo's (3-5 v. 5-3). As predicted by the AW method I am using, Myogiryu won the match. So in selected matches, the AW metric is 1-0.

DAY 2 PREDICTIONS

Day 2 (day 10 of the basho) offers 2 matches in which the AW total is at odds with the W-L records of the rikishi:

Ichinojo (5-4, 6AW) vs. Tochiozan (4-5, 12AW)

Hokutofuji (5-4, 8AW) vs.  Myogiryu (4-5, 14AW)

So AW says take the two rikishi with 4-5 records.

PROSPECT

Again, I do not expect this metric to be particularly beneficial on a match-to-match, entire banzuke-wide level. On such a basis, sumo is wildly unpredictable. On day 9, Ryuden (2-6, 0AW) defeated Kaisei (7-1, 12AW). Show me the metric that predicted that, and also give me an invitation to visit you on your yacht in the Mediterranean. My expectation is that this might help pick out one or two matches per day where the records of the rikishi are not inductive of what to expect in the match.

Beyond including the measurement of accumulated losses, I am wondering if AW should be static or dynamic. Should wins accumulate to a wrestler only on the day he wins, or should wins accumulate throughout the tournament. For example, Nishikigi is 4-5 but only has one AW because on day 3 he defeated Kakuryu who was 1-1 at the time. However, on day 1, Nishikigi defeated Goiedo who has now won 3 matches. Kakuryu won an additional match on day 4. So given the wins of wrestlers when he defeated them, Nishikigi only has 1AW, but given all the wins of the defeated wrestlers throughout the tournament, he has 5AW. Taking into consideration cumulative AW might be more helpful. But since sumo is a zero-sum game the line might have to be drawn there, otherwise we will just be predicting ties for very match!

[Pandaazuma]

Perfect thread for @Randomitsuki

[Yamanashi]

Sorry to get all meta- here, but it's important to remember that any truly 100% predictive metric will mean there's nothing to argue about.  That would be the death of this Forum, and the end of fandom in general.  So, beware what you wish for.

 

[BroadMeadow]

DAY 2 (basho day 10) REPORT

Yesterday was a pretty bad one for the AW metric. An overall record of 8-9-2 and neither of the predicted upsets materialized. That makes the 2-day total: 17-16-5 and 1-2

DAY 3 PREDICTIONS

Matches with even W-L records with predicted AW winner:

• Kotoeko over Sadanoumi
• Daieisho over Kotoyuki
• Yutakayama over Onosho

Matches where lower W-L is predicted winner:

• Asanoyama over Yago
• Chiyotairyu over Kaisei
• Takayasu over Okinoumi

By ranking, Takayasu over Okinoumi can't be considered upset per se, but Okinoumi does have the better W-L record this basho.

[BroadMeadow]

DAY 3 (Basho 11) REPORT

Banzuke : 7-10-2 (total: 24-26-7)

Highlight matches- 3-3-1 (total: 4-5-1)

DAY 4 PREDICTIONS:

Predicted winners in tied W-L record matches-

·         Takayasu

·         Hokutofuji

·         Shodai

·         Ikioi

·         Asanoyama

Predicted winners with worse W-L record than opponent:

·         Takarafuji

[BroadMeadow]

DAY 12 REPORT

Banzuke- 13-5 (37-31-7)

Highlight matches- 5-1 (9-6-1)

DAY 13 Predictions:

Predicted winners in tied W-L record matches:

• Onosho
• Chiyotairyu
• Asanoyama
• Shohozan

Predicted winners with worse W-L record than opponent:

• Chiyoshoma

[BroadMeadow]

DAY 13 REPORT

Banzuke 9-9-1 (46-40-8)

Highlight matches 2-3 (11-9-1)

DAY 14 PREDICTIONS

Predicted winners in tied W-L record matches:

• Kotoshogiku

Predicted winners with worse W-L record than opponent:

• Kagayaki

[BroadMeadow]

DAY 14 REPORT

Banzuke- 12-4-2 (58-44-10)

Highlight- 0-2 (11-11-1)

DAY 15 PREDICTIONS:

Predicted winners in tied W-L matches-

• Shodai
• Tochiozan

Predicted winners with worse W-L record than opponent:

• Myogiryu

[Pandaazuma]

It's a lot more fun if you just join the games at that link I provided above! Hope to see you in there next basho!