The first posts about this topic:
Baseball Betting System #1
Baseball Betting Tactics #2

The Baseball Underdog system is based on the law of average numbers, showing that during a long period of time MLB Underdog teams win favorites in 4 games from 9 (or 44%). But even if you bet an underdog on the moneyline, you will earn more, as the odds is +130, you bet 10$ to win 13$, it is much better than 5 wins and 4 losses of the favorites every day.

For example, let’s suppose that you stake 10$ on the every underdog today, which coefficient is +130 on the moneyline. If you win 4 games, you gain 13$ for every correctly guessed match, or 52$ in sum. But you lose 50$ because of 5favorites’ wins.

But instead of betting all the underdogs every day, you have to choose games by following these three simple criteria:

First of all: remove the games, in which the underdog has lost three or more times in succession, and/or the favorite has won three or more times in succession.

It will prevent you from betting on “bad”, “tricky” underdogs, which meet the excited teams, whose bettors can “thrash” the pitchers of this team.

Secondly: remove the games, moneyline on underdogs of which is higher than +150, or something about it.

If the odds are +150, it is almost doubtless that underdog has broken its set. So cross out such games and go on.

Thirdly: remove the games, in which underdog plays against one of the pitchers, who is, according to the earned-run average (ERA), in Top20 of the league.

Nowadays there are many newspapers on internet web sites, which can provide you with ERA statistics, but I insistently recommend you to use the “standardized” ERA ratings, which are calculated by Jeff Sagarin (R), and situated on the web site USA Today’s.
http://www.usatoday.com/sports/sagarin.htm

In this rating Sagarin uses some complicated data, to provide every pitcher in the league with a number that shows his “real” ERA. He identifies ERA for every pitcher in comparison with the actions of other pitchers from 1946 to 1999, and according to these comparisons he calculates NPERA (or “normalized predicted earned-run average”).