Timecontrol, Win- & DrawRatio

[Edmund]

Today I made a strange oberservation consulting the CEGT and CCRL chess engine rating lists.

Comparing the different timecontrols I found in both rating lists that with increasing timecontrols the draw rates increased too. That was nothing new to me and no surprise.

Timecontrol - Draw-rate (on CEGT)
40/4 - 30.9%
40/20 - 35.6%
40/120 - 41.3%
40/120 (4cpu) - 45.2%

But then I also compared the win ratio for white and found out that it was staying equal or even increasing slowly.

Timecontrol - Win Ratio for white
40/4 - 55.0%
40/20 - 54.6%
40/120 - 55.4%
40/120 (4cpu) - 55.9%

I would have assumed they would decrease and tend to reach 50% as draw rate continues to rise, but this is the contrary. The data would suggest that the likelihood for white to win increases the higher the level of play.

Any interpretations? Is this a sign of the game of chess being ultimately won for white? Did anyone make any similar observations or tests?

regards,
Edmund

[swami]

Any interpretations? Is this a sign of the game of chess being ultimately won for white? Did anyone make any similar observations or tests?

regards,
Edmund

Did you ignore draw games when you calculated the win percentages for white?

If so, win % for black would be about 45%

Player with White has the first initiative because he develops and activates his pieces first. But since chess is so deep, the probability for a draw is more.

[Edmund]

Any interpretations? Is this a sign of the game of chess being ultimately won for white? Did anyone make any similar observations or tests?

regards,
Edmund

Did you ignore draw games when you calculated the win percentages for white?

If so, win % for black would be about 45%

You are correct: "my" win-ratio = (wins + draws/2) / N

Here is the full report from the CEGT 40/4 tournament:

Games : 582838 (finished)

White Wins : 230392 (39.5 %)
Black Wins : 172431 (29.6 %)
Draws : 180015 (30.9 %)
Unfinished : 0

White Perf. : 55.0 %
Black Perf. : 45.0 %

Player with White has the first initiative because he develops and activates his pieces first. But since chess is so deep, the probability for a draw is more.

I would have thought so too, but these testreports would suggest the opposite.

[Kirill Kryukov]

Today I made a strange oberservation consulting the CEGT and CCRL chess engine rating lists.

Comparing the different timecontrols I found in both rating lists that with increasing timecontrols the draw rates increased too. That was nothing new to me and no surprise.

Timecontrol - Draw-rate (on CEGT)
40/4 - 30.9%
40/20 - 35.6%
40/120 - 41.3%
40/120 (4cpu) - 45.2%

Hi Edmund,

For what it's worth, here I have some draw rate statistics for my tournament: http://kirill-kryukov.com/chess/kcec/draw_rate.html

Basic result is that draw rate in a match depends quite a lot on rating difference between the two opponents, and also on their average rating.

When you compare one game database to another (like CEGT to CCRL, or CCRL 40/40 to CCRL 40/4), you have to consider the possibility that the average pairing might differ between the two databases, resulting in different overall draw rate.

While I agree that slow time control is expected to produce more draws, another possible factor is: With slow time control testers only have time for the strongest engines, which make more draws averagely. With faster time control the testers can afford to test larger number of engines, so weaker engines are also participating, reducing the overall draw rate.

I don't know how/if this kind of logic also applies to white advantage, but I guess there may exist some dependence on pairing there too.

Best,
Kirill

[Edmund]

Hi Edmund,

For what it's worth, here I have some draw rate statistics for my tournament: http://kirill-kryukov.com/chess/kcec/draw_rate.html

Basic result is that draw rate in a match depends quite a lot on rating difference between the two opponents, and also on their average rating.

When you compare one game database to another (like CEGT to CCRL, or CCRL 40/40 to CCRL 40/4), you have to consider the possibility that the average pairing might differ between the two databases, resulting in different overall draw rate.

While I agree that slow time control is expected to produce more draws, another possible factor is: With slow time control testers only have time for the strongest engines, which make more draws averagely. With faster time control the testers can afford to test larger number of engines, so weaker engines are also participating, reducing the overall draw rate.

I don't know how/if this kind of logic also applies to white advantage, but I guess there may exist some dependence on pairing there too.

Best,
Kirill

Thanks Kirill, your tournament webpage is as always an interesting read.

Concerning the drawrate increase with average strength. I know from my tests where I play very rapid games between different versions of Glass that with increasing timecontrol the drawrate goes up significantly. Its something like:
10s/game => 22.5%
20s/game => 33.6%
60s/game => 36.5%

so your second graph very much confirms this.


Concerning the increased win_ratio with longer timecontrols, I didn't consider the problematic that not the same engines are tested under the longer timecontrols. But if you are right and on average stronger engines are tested on longer timecontrols, this doesn't weaken the argument. stronger engines should lead to even higher draw rates and thus smaller win_ratios, but the opposite is happening.

regards,
Edmund

[UncombedCoconut]

But then I also compared the win ratio for white and found out that it was staying equal or even increasing slowly.

I would have assumed they would decrease and tend to reach 50% as draw rate continues to rise, but this is the contrary. The data would suggest that the likelihood for white to win increases the higher the level of play.

Any interpretations? Is this a sign of the game of chess being ultimately won for white? Did anyone make any similar observations or tests?

I don't see the problem: the % of games won by white (not counting half-points) is decreasing, same for black. Note also that "your" win ratio is equal to (#white wins - # black wins + N)/2. If you think of a TC increase as tending to turn won/lost games into draws, it increases "your" win ratio as long as it lets white eke out more draws than black. Maybe up to a certain point this is true (because the side with first-move advantage has an easier time avoiding losses through more precise play), after which it becomes false (because black has more lost games to start drawing)?

[Edmund]

But then I also compared the win ratio for white and found out that it was staying equal or even increasing slowly.

I would have assumed they would decrease and tend to reach 50% as draw rate continues to rise, but this is the contrary. The data would suggest that the likelihood for white to win increases the higher the level of play.

Any interpretations? Is this a sign of the game of chess being ultimately won for white? Did anyone make any similar observations or tests?

I don't see the problem: the % of games won by white (not counting half-points) is decreasing, same for black. Note also that "your" win ratio is equal to (#white wins - # black wins + N)/2. If you think of a TC increase as tending to turn won/lost games into draws, it increases "your" win ratio as long as it lets white eke out more draws than black. Maybe up to a certain point this is true (because the side with first-move advantage has an easier time avoiding losses through more precise play), after which it becomes false (because black has more lost games to start drawing)?

My reasoning was similar to your final point. Having no prior information about how the increased draws are distributed I would have assumed an equal distribution and as #white wins > #black wins, the likelihood of a white win turning into a draw would be higher with a lower win ratio as a consequence.

because the side with first-move advantage has an easier time avoiding losses through more precise play

As you increase the level of play both sides will avoid certain losses, finding a rescue move beyond the horizon of faster time controls. The question is now, why does white rescue more losses than black, or why does white turn more drawn games into wins than black at longer time control. Just the "first-move advantage" on its own is not the solution. Having the first move is a significant advantage and this regardless of the the strength level. The main point is why does the first-move advantage increase with playing strength?

regards,
Edmund