Accumulated wins as an in-basho performance metric

[yorikiried by fate 2,028]

@BroadMeadow

I have trouble interpreting your daily shorthand posts without reviewing the whole concept (I feel lazy). Can you summarize your own impression on how this worked out?

[BroadMeadow 43]

DAY 15 REPORT

• Banzuke- 5-12-1 (63-56-11 52.9%)
• Highlight matches- 3-1 (14-12-1 53.8%)

2 hours ago, yorikiried by fate said:

@BroadMeadow

I have trouble interpreting your daily shorthand posts without reviewing the whole concept (I feel lazy). Can you summarize your own impression on how this worked out?

As for the everyday reports, the daily results are given first and then the cumulative results are given in the parenthesis. The "ties" are matches where the rikishi had the same AW total or matches with a fusen. The "highlight matches" include two types of matches- matches between rikishi with the same W-L records, and matches in which the rikishi with the lower/worse W-L is favored to win by the AW measurement. For this basho, accumulated wins are figured by adding any wins of the defeated wrestler to the wins of the victorious wrestler.

How did it work? I am working on my impressions. My hypothesis was that the metric would be helpful in picking "upsets" (with upsets based purely on W-L records). According to this site: https://www.thesportsgeek.com/sports-betting/math/ you have to win 52.4% of your bets to break even. Since the metric predicted winners at a 53.8% rate, I would have made money if I had actually wagered it. But this is for entertainment purposes only...

I do not know whether to be concerned or happy that the highlight match win percentage basically matched the entire banzuke percentage. The banzuke record is basically just a scratch record- against the W-L of the wrestlers. So for this basho at least, the metric picked "upset" winners at the same winning rate as someone who was just picking against the W-L records.

There are some things that could be added or changed to potentially improve the metric:

• making the wins truly cumulative. Right now if someone beats a 1-0 Hakuho he gets 1 accumulated win. But if someone beats a 10-1 Hakuho, he gets 10 accumulated wins. Both wrestlers defeated Hakuho, but one's victory was 10x more valuable than the other. I have already started work on the spreadsheet to include this in March.
• adding accumulated losses to metric. This is in the March spreadsheet too.
• figuring in the historic record between the 2 rikishi. It would seem to matter if one wrestler "has the number" of another wrestler.
• figuring in some sort of calculus for the difference in rank. A 9-3 Yokozuna should probably be favored over a 12-0 M16 no matter what the other numbers say. (I know, I know, a Yokozuna would never face an M16 on day 13 of the basho, but you get the point.)

In short, if I was trying to make money off of this, I would have this time. But I would like to be a little more confident about the next time! I jumped into this without a whole lot of forethought so it is a work in progress. Nothing is ever going to be 100% accurate, but if I could get up to 60% I would feel really good about it.

[BroadMeadow 43]

On ‎26‎/‎01‎/‎2019 at 20:57, Pandaazuma said:

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!

I got you! I am slow to make commitments, but I have the tab open and I will spend time between now and March getting familiar with everything. Hopefully I will join in the fun then. Keep hounding me though!

[Pandaazuma 1,310]

17 minutes ago, BroadMeadow said:

I got you! I am slow to make commitments, but I have the tab open and I will spend time between now and March getting familiar with everything. Hopefully I will join in the fun then. Keep hounding me though!

Great. I guarantee you'll love it. A real challenge for anyone into prediction.

[Shinrei 47]

10 hours ago, BroadMeadow said:

There are some things that could be added or changed to potentially improve the metric:

• making the wins truly cumulative. Right now if someone beats a 1-0 Hakuho he gets 1 accumulated win. But if someone beats a 10-1 Hakuho, he gets 10 accumulated wins. Both wrestlers defeated Hakuho, but one's victory was 10x more valuable than the other. I have already started work on the spreadsheet to include this in March.
• adding accumulated losses to metric. This is in the March spreadsheet too.
• figuring in the historic record between the 2 rikishi. It would seem to matter if one wrestler "has the number" of another wrestler.

If you're open to input on this... I have a few ideas that may be worthy of your consideration.

• Include current record difference as well. Rikishi A (with a record of 1-7) could/should get more credit for defeating Rikishi B (7-1) than Rik-C (3-4) defeating Rik-D (4-3).
• Scale and limit the accumulated wins/losses to a range of about ± floor(current day/2).
• Limit the historic record to 3 matches within the past year with decreasing importance (such as 3 bits for the previous, 2 bits for the one before that, 1 bits for the last).
• Math aside, if a rikishi is hurt/injured/shaken/etc from a bout and doesn't withdraw, that could decrease their chance of victory in their next bout

Comments?