Musings on Chess Openings

In a recent episode of the Perpetual Chess Podcast, Ben talked about how different openings perform at different levels:

It had me looking at how different openings score. I probably I should be doing this anyway, but out of curiosity and not going deep into the variations: 1. e4 how do the different moves look, and then like sorting by every 200 point rating band […] and it's funny the double king pawn at the GM level it wins less, but draws more and overall black's expected value is higher, but Sicilians do still win more, but once you get to the club level, 1. e4 e5 is the worst opening, even though it's objectively the best, …

As someone who plays 1. e4 e5, this had me worried. I’d heard something similar before, and as someone who always picks their openings based on how it scores at Master level I felt very seen.

Since the data was easily available, I decided to take a look. First at opening moves (1. e4, d4, c4, Nf3) and then some variations.

The first thing I looked at was expected score (%wins + 0.5 * %draws).

This suggested that:

1. 1. e4 is ~worst at ~all levels on Lichess

2. This is still true(!) at master level

I couldn’t really believe that e4 was quite this bad, and a friend of mine pointed out that these results could be being confounded in 2 ways.

1. Who is playing the opening (could their be a bias in the average rating in each bucket)

2. Who the opening is played against? (eg Do masters selectively play c4 / Nf3 against weaker opponents?)

One way to account for this is to look at the Performance Rating of each opening. This is [average opponent] + 400 * (average_score - 0.5). We can then compare this to the rating of the player playing the opening to decide whether it is “good” relativel to their level or not.

To make this really concrete in the Blitz 2000-2200 rating range on Lichess, the average rating playing:

1. e4 - 2077 against 2079

2. d4 - 2082 against 2079

3. c4 - 2087 against 2079

4. Nf3 - 2091 against 2078

So we shoud expect c4 to outperform e4, just because it’s stronger players playing it within a given rating range.

When we adjust for this, we still find that d4/c4/Nf3 outperform e4, but the difference isn’t so stark:

It’s interesting to see that 1. e4 reclaims it’s “Best by test” title among masters.

As an 1. e4 player, this definitely has me questioning whether or not I should consider a switch.

I tried to determine when the switch-over occurs for e4 vs d4 at master level, by looking at the performance ratings in the ChessTempo database, but e4/d4 were ~indistinguishable for all master levels:

The next thing I looked at was the same analysis for 1.e4 defenses. (If there are other ones you are interested in, please let me know).

Lessons learned here:

1. The Alekhine is extremely good for club players, potentially even at master level (although perhaps there are so few games played by masters because they know how to refute it, and therefore the sample is biased)

2. 1.e4 e5 is as bad as Ben says until a very high level. Need to be above ~2400 Lichess Blitz before it’s comparable to other 1.e4 replies.

3. The Sicilian is strong at every level.

Since I recently switched from the Italian to the Ruy Lopez, I checked the relative performances for those variations within 1.e4 e5:

At least it seems like I made a good change here.

I also checked what I should be playing against the Sicilian, since I play both Alapin and Open, it seemed worthwhile checking on their performances:

It seems like the data is strongly encouraging me to focus more on the Alapin. I was curious to understand where this breaks down at the elite level (why don’t top players play it more if it scores so well?)

The answer seems to be that > 2500, black seems to have found an edge.

If I am to conclude that 1. …e5 should no longer be my reply to 1.e4, what should I do instead. The section above seems to suggest the Sicilian, if so - which one?

(NB: lower is better here, since we want Black outscoring white). At ~club levels, the Dragon seems to do very well, although it suffers badly as ratings progress beyond 2200. The Najdorf, Classical and Sveshnikov all have solid performances across all levels, so I will do some more research into those.

Finally, I looked at replies to 1.d4. I’ve never played d4, so I don’t really have a good sense of whether I’ve picked representative openings. Please let me know if there’s others you’d like to see:

I guess the things which stand out to me are:

1. Benko/Benoni are very playable at club level

2. Grunfeld is strong at every level