Why Engines in CCCC tend evaluate the position better for themselves.

 I  have been following the final games between Stockfish and Houdini.

(https://www.chess.com/computer-chess-championship)

Looking at the evaluation graph for both black and white, it is almost always the case that the engine playing white evaluates the position value to be greater (more positive) than the evaluation of the engine playing black.

Why does an engine A think it is performing better than what its opponent thinks engine A is performing, and vice versa?

Yes. It seems that engines are designed to play for a win if they are white and to play for a draw with black. Engines, after all, are programmed by humans. So does that mean that we humans, like engines, do feel more confident than what our opponent thinks we feel?

Stockfish have contempt + 0.21since around ?March 2018.  It means SF will evaluate an equal position as +0.21 instead of 0.00

Houdini and Komodo have contempt as well, but they are closed source. I cant see the code.

It surely is an interesting thing that every engine has its own philosophy of giving weights. This explains the different evaluations the two engines give. However, I think this does not explain why every time SF is playing white, its evaluation is more positive then the evaluation of Houdini. And when SF is playing black, its evaluation is more negative the that of Houdini. 

I don't know how this exactly works. But I believe this also does not explain the observation. When SF is playing white, it gives more points to white than what it gives to white when it'ss playing black.

I totally agree with you. It makes sense that every engine evaluates differently, and the better the engine the more accurate it is. However, this is not what I was talking about. I think I explained very baldly. My question is about the weird tendency of engines to like its own side more than its opponent. Let me explain in more a detailed example.

Let's take 50 drawn games. 25 of which had SF playing white. In those games that had SF playing white, say the average evaluation SF gives for the first 20 moves was +0.25, then say black could equalize afterwards. If we looked at the other 25 games, which had SF playing black, we notice that the average evaluation of the first 20 moves is less then +0.25, it was about 30% less, so +0.18 ( this can mean that SF played with white better then Houdini played with white).

Now if we do the same but for Houdini. Say the average evaluation of the first 20 moves for the games Houdini played white was +0.23 (close but not equal to the evaluation SF gave earlier +0.25 as the two engines use different algorithms). If we looked at the rest of the 25 games( Houdini played with the black pieces), we would see that the evaluation Houdini gives roughly +0.16. (This suggests that Houdini played better with white then SF did.) 

This gives a contradiction. SF thinks it played better. And Houdini also thinks it played better. So there is something weird.

I hope the following sample 6 games and graphs from the CCCC help further clear up my point. 

It is easy to see. Most engines give half move advantage for white around + 0.20.

SF playing white and assume 1. e4 as +0.20+0.21= +.0.41

In the same position if SF play as black he will see -0.20+0.21= +0.01 from black.

In total, there is a huge difference in evaluation in changing side =  0.42 difference due to contempt.

I don't see why SF would subtract 0.20 if it is playing black. The contempt should give the half move advantage to white not black. So it would add 0.20 in either way, not subtract, I believe.

Contempt is added to Stockfish point of view to avoid draw especially as black. In short, without contempt, SF in playing as black see white pieces  relatively more valuable. SF in black might see his Q as 9.99 relative value ( because of less mobility in black) and see white queen as 10.00. So SF as black will trade down Queen. And also attempt to trade all pieces.

Final result ended as draw . But we dont want 3500 rated SF in black to draw vs 3200 engine playing white.

That is main reason contempt was added.

With contempt SF as black might see his Q as 10.01 and opponent queen as 10.00. So he keep his pieces.

As chess is too drawish, exchanging pieces lead to more draw. To win a game, you must keep your pieces as much as you can until you see decisive advantage( or opponent mistake).

No, it perfectly explains it. That's exactly what contempt does. SF is biased towards the color it's playing as (or the player whose move it is if you're doing analysis)

Chess isn't that drawish, it's just engines are too dumb to understand long term ideas like keeping pieces on the board gives more winning chances.

For example I played SF at knight odds to see how it plays, and of course it plays like an idiot, trading everything and letting me win the endgame without any effort.