Stats and gaming strategies

[Kotonosato]

May I ask what everyone's specialty is? It's clear that you're all very familiar with statistics. Assuming you're not actually statisticians, what kind of research do you do?

We may not be statisticians, but we are Germans which is close enough if you look at various Sumo Games. (Laughing...)

I don't follow the sumo games, but I'm curious:

Does someone with a thorough knowledge of statistics win more often than not? Or are you foiled by chance events just as often as the rest of us?

Do you tend to make predictions based completely on the stats, or are you constantly tweaking your predictions based on intuitions and observations (e.g. "that rikishi looked so good last basho, he is bound to win more often than the stats suggest").

Can you make money gambling?

[Doitsuyama]

Does someone with a thorough knowledge of statistics win more often than not? Or are you foiled by chance events just as often as the rest of us?

I'd say using statistics is a slight advantage, nothing more.

Do you tend to make predictions based completely on the stats, or are you constantly tweaking your predictions based on intuitions and observations (e.g. "that rikishi looked so good last basho, he is bound to win more often than the stats suggest").

What IS "the stats"? Think about it and tell me what "the stats" are meant to be in this context. How would you use "the stats" to make picks? Maybe you agree then that even thinking about that can give helpful insights.

[Kotonosato]

Do you tend to make predictions based completely on the stats, or are you constantly tweaking your predictions based on intuitions and observations (e.g. "that rikishi looked so good last basho, he is bound to win more often than the stats suggest").

What IS "the stats"? Think about it and tell me what "the stats" are meant to be in this context. How would you use "the stats" to make picks? Maybe you agree then that even thinking about that can give helpful insights.

I'm afraid the only insight this might reveal is my complete ignorance of statistics ( (Laughing...) ), but that hasn't stopped me yet, so I'll continue...

For example, what if you were trying to "guess the banzuke"? Based on prior outcomes, you might have calculated that a rikishi with a 9-6 record at East Maegashira 12 is typically raised 2.5 positions for the next basho. On the other hand, due to his popularity with the crowd and the kyokai, you might feel that he will be given a slight advantage, so instead you predict that he will be given a 3 or 3.5 position boost.

I suppose you could quantify "popularity" into the equation, too, or would you just leave the final decision up to "intuitive tweaking"?

Another example: What if you were trying to predict the final results of a given rikishi before the start of a basho? You know that he has been raised five positions on the banzuke since last basho, so you can calculate his average score after a five-rank promotion. You can also predict what his score should be at his new rank based on previous results. Maybe you can even create a formula to combine these two and give you an accurate reading of the two factors together.

My question is: Would you leave it at that, or would you go beyond "the stats" and say, for example, "regardless of his past performances, I think he is approaching the end of his career. He will not perform up to what 'the stats' predict." Thus, you would conclude that he will be 6-9, rather than, say, 7-8 as the stats might have predicted?

Finally, (without making me look too stupid, please) is that the kind of thinking that you bring to bear on the sumo games? Or is your approach completely different?

[Doitsuyama]

Finally, (without making me look too stupid, please) is that the kind of thinking that you bring to bear on the sumo games? Or is your approach completely different?

Well, I think your examples show that there are quite different possibilities of using "the stats" and there still is enough brain work involved to make the picks individual. By the way, I am not convinced your suggestions would lead to good picks, and I would say my approach is different.

[Asashosakari]

My question is: Would you leave it at that, or would you go beyond "the stats" and say, for example, "regardless of his past performances, I think he is approaching the end of his career. He will not perform up to what 'the stats' predict." Thus, you would conclude that he will be 6-9, rather than, say, 7-8 as the stats might have predicted?

Well, recognizing and incorporating typical aging patterns is also "stats", in a way. However, I for one only resort to that when it comes to major events, like expecting somebody to completely fall off a cliff, or concluding that some prior result was a mirage and should be largely ignored for future predictions (Iwakiyama's 12-3 in Juryo is an excellent candidate for that treatment next basho). But predicting 6-9 or 7-8 for a rikishi...that's just way too fine a distinction and pretty much a toss-up anyway for each individual case. AFAIK, some players go right by the numbers for that kind of fine-tuning (whatever that might mean for a particular player, since everyone of course has their own guidelines and assumptions), some do it purely on gut feeling, and then there's everything in between.

Where stats can help you is in not missing the "target" altogether (like expecting Tokitsuumi to go 12-3 next basho simply because he's falling far down the banzuke), so I pretty much agree with Doitsuyama that it's only a small advantage. And it's not strictly necessary to use numbers...well, for me it is, because I have negative intuition, if anything, and without extensive consideration of numbers I would absolutely suck. (I am not worthy...) Other players have a much more intuitive feeling for sumo, especially as far as daily games are concerned.

Edit: Perhaps one more (minor) way that a math/stats background can help is via introspection...e.g. if you have 10 rikishi that you (correctly) expect to reach a small make-koshi, it's probably not optimal if you invariably expect all of them to get 7-8 or all 6-9. In other words, some math knowledge may help prevent falling into negative habits while making picks.

Edited August 5, 2007 by Asashosakari

[Randomitsuki]

My question is: Would you leave it at that, or would you go beyond "the stats" and say, for example, "regardless of his past performances, I think he is approaching the end of his career. He will not perform up to what 'the stats' predict." Thus, you would conclude that he will be 6-9, rather than, say, 7-8 as the stats might have predicted?

Finally, (without making me look too stupid, please) is that the kind of thinking that you bring to bear on the sumo games? Or is your approach completely different?

My interest in sumo games is focusing on numbers exclusively, a perverse and self-afflicted code of honor that results in trying to develop stats for everything. I believe that even gut feeling is just the application of "hidden stats", so I seek ways to translate gut feeling into numbers. Two examples:

1) I take aging (and/or streaks of disappointing bashos) into account numerically. This backfired last basho with Iwakiyama. While the "raw numbers" clearly suggested a double-digit number of wins, I used a formula that applies different algorithms for rikishi that are supposedly on a decline. These alternative stats resulted in a 7-8 prediction which, of course, was as wrong as could be.

2) Most of my stats last basho indicated that Wakanosato would have a strong basho. My gut feeling said that this is complete nonsense. I sticked with the numbers, and once again, the gut decision would have been so much better.

The conclusions that I draw from such wrong decisions, however, is always the same: my stats simply aren't good enough yet, and they need additional tweaking (I am not worthy...).

[Kotonosato]

It's interesting to hear how everybody thinks.

By the way, I am not convinced your suggestions would lead to good picks, and I would say my approach is different.

Without giving away your "secret formula", can you tell us a little about your approach, too?

[Itachi]

Wow, what an interesting topic for number loving fans in the gap between basho that usually has us starving for fresh data.

This thread should probably be sent off to the games sub-forum now as the discussion about how some of us use the numerical data to play the games.

As for myself (not German but I did finish a Math degree) I like looking at rikishi data and I use is to help inform my decisions in sumogames but I don't calculate any statistics based on them. In particular, I like to look at rikishi results over time and look for trends and patterns in the way a technical analyst might look at stock charts.

Supporting details such as rikishi age, size, and known health/injury concerns I use to adjust my expectations.

As far as distributions of records, streaks and such, I prefer to go by gut feeling guided by observation of recent trends. It may be that those who use statistical analysis effectively can capture and quantify all the factors I consider and have a more consistent and efficient system for making predictions but I haven't been motivated to develop such a system myself.

Also, I am concerned about the 'Law of large numbers' issue. I'm not convinced that we have enough data about each type of rikishi in each situation to be able to make reliable statistical conclusions. I call this concern of mine in sports predicition contests the 'Law of small numbers'.

My 'Law of small numbers' states that if there isn't enough historical data over a time period short enough to be relevant then conclusions based on statistical analysis of that data is not very useful in predicting future outcomes.

[Randomitsuki]

What I love about sumo gaming is its inherent unpredictability. Someone who does not have the slightest clue about sumo will have a prediction performance of 50% in Sekitori-Toto. Now I have just checked Norizo's (arguably the world's leading daily gamer) prediction performance in Sekitori-Toto over the last six basho. Guess what? Even Norizo who is lightyears ahead of most of us, has predicted only 59.4 percent of sekitori bouts correctly. And that includes fusensho wins, yokozuna bouts etc.

That the difference between 50% and 59% covers a whole universe of playing ability is the most fascinating aspect for me.

It's probably stupid to think that an additional per cent or so can be gained by using more and more stats. But I won't give up...

[Doitsuyama]

Without giving away your "secret formula", can you tell us a little about your approach, too?

I think I have written about that before, it really isn't much of a secret. Basically I assign strength ratings to each rikishi and continue from there. Strength ratings were discussed before as well, and I believe there are a few more playing with such an approach.

[Asashosakari]

I believe that even gut feeling is just the application of "hidden stats", so I seek ways to translate gut feeling into numbers.

I think I agree with that take, though I don't feel the urge to codify it for my own gaming. (A sekitori...)

1) I take aging (and/or streaks of disappointing bashos) into account numerically. This backfired last basho with Iwakiyama. While the "raw numbers" clearly suggested a double-digit number of wins, I used a formula that applies different algorithms for rikishi that are supposedly on a decline. These alternative stats resulted in a 7-8 prediction which, of course, was as wrong as could be.

You failed to consider the "the first demotion to Juryo will always spark a veteran guy's performance temporarily" exception... (I am not worthy...)

What I love about sumo gaming is its inherent unpredictability. Someone who does not have the slightest clue about sumo will have a prediction performance of 50% in Sekitori-Toto. Now I have just checked Norizo's (arguably the world's leading daily gamer) prediction performance in Sekitori-Toto over the last six basho. Guess what? Even Norizo who is lightyears ahead of most of us, has predicted only 59.4 percent of sekitori bouts correctly. And that includes fusensho wins, yokozuna bouts etc.

That the difference between 50% and 59% covers a whole universe of playing ability is the most fascinating aspect for me.

It's probably stupid to think that an additional per cent or so can be gained by using more and more stats. But I won't give up...

Old-ish Sekitoto stats

Edited August 6, 2007 by Asashosakari

[Kotonosato]

Someone posted a dissertation about sumo at the address below. It includes a statistical analysis of bout rigging.

http://homepage.sunrise.ch/mysunrise/mvonw...issertation.pdf

I'd like to hear people's thoughts on its validity.

[Randomitsuki]

Someone posted a dissertation about sumo at the address below. It includes a statistical analysis of bout rigging.

http://homepage.sunrise.ch/mysunrise/mvonw...issertation.pdf

I'd like to hear people's thoughts on its validity.

I've already stated my opinion on SML. Certainly an interesting piece of work using some novel approaches to determine bout rigging. The validity of the data appears to be high, there are lots of inexplicable patterns happening in bouts. The problematic thing, however, is the validity of the conclusions. As discussed elsewhere, many of the strange patterns might as well be explained away by the assumption that rikishi on the verge of KK try a little harder. You don't need the exchange of money for that. That is not to say, of course, that yaocho does not exist. It's just that we can hardly know from the numbers themselves, so other accounts (confessions, proof of transactions being made etc.) would be much harder evidence.