Making a Correspondence Opening Book

    A few years back ICCF IM (current rating 2454) Wolf Morrow (aka FirebrandX) wrote an article for chessdotcom on opening preparation which you can read HERE. 
     In the article Morrow stated that on ICCF (and LSS) deep engine analysis drives most of your opponents' moves so it is vital to be well-prepared in the opening. His method is to create a database of all ICCF games then extract the first 40 moves into a new opening book. In the article Morrow also pointed out that many top GMs build their opening innovations based on ICCF games.
     I took Morrow's advice and in addition to ICCF games I also included 15 years of games played on LSS and then merged them all into one large database. My next step was to delete all the games played by players rated under 2200. I also chose to use only games played in the last four years because I reasoned that prior to that engines were not nearly as strong as they are today and I didn't want old games with outdated opening analysis included. This gave me a database of 91,000 master level games from which to create my opening book. The end result was an opening book with over 6 million positions!

Some statistics:
Opening moves by popularity. In parentheses is the won, lost, drawn percentage.

1.e4 (31-21-49)
1.d4 (29-19-51)
1.c4 (29-22-50)
1.Nf3 (27-21-52)

In reply to 1.e4 the Sicilian was the most popular (26-21-53) followed by 1...e5 (32-19-49) and the French (34-21-45). Against 1.d4 black gets his best results with 1...Nf6 (27-20-53) and 1...d5 (30-18-52) the King's Indian (29-18-53) and the Nimzo-Indian (23-20-57).

     Just for fun I made a book using all MY games. 1.e4 percentages were +32 -31 =37 while 1.d4 percentages were +39 -28 =33 so there was not much difference at the amateur level and I see I played the Grob Attack (1.g4) quite a few times and my percentage with it (I have never faced it) has been +86 -0 =14. That does not mean I didn't lose any games playing it, just that the loss percentage was less than one percent! I hasten to add that almost all of those games were against fairly low rated players which explains why it was so successful.
     In the end though I'm not sure how important this is at any level below the correspondence GM level...and by that I mean the really big boys that play at the world championship level.
     In one recent game against an ICCF rated 2300+ opponent as black I met his 1.Nf3 with 1...a6 and drew. In another game as white I played the Caro-Kann Milner-Barry Gambit against an OTB FM and drew.
     There was a surprise in one of my games. Against an ICCF Senior IM with me as white we opened with 1.d4 d5 2.c4 dxc4 and now I played 3.e4 which according to my new opening book scores (percentages) +40 -16 =44!! What surprised me was the results of meeting the QGA with 3.e4. Upon checking I found that in the Rybka2 opening book the percentages are +41 -33 =26 and the Fritz opening book shows similar results. The meaning is that 3.e4 scores quite well for white and in correspondence play he has slim losing chances (16 percent vs. 33 percent OTB). 

Additional information on engine opening books:
Chess Programming Wiki 
Good Opening Books for Arena 
Free Hiarcs Opening Book Downloads 
Beginners Guide to Building Opening Books