See how we're doing so far this year (by day) (by team).
Why Use Our Method?
Past Experience
The approach used has been tested during the past four baseball
seasons with encouraging results. We have had positive outcomes each
year.
Why the method makes sense
You might be wondering why we think our method could do this when
our expertise is in mathematics and not in baseball. Those that play
poker games, scoffed at mathematicians as well until math geeks
started winning major poker titles. Others like to play games that can be analyzed mathematically,
like Jackpotjoy.com's 90 ball bingo online game.
In any case, when the
payoffs are set each day for each team in each contest, these
payoffs are based on the amounts bet on each team by bettors. We do
not suppose that we know any more than baseball insiders. However,
in order to win in the long run, we just need to be better than the
average bettor. (One can use the same analogy in playing the stock
market. In order to do better than the averages, one does not need
to have any inside knowledge. One just needs to understand more
than the average investor.)
How our method works
We have a copyrighted mathematical model for analyzing the
probability a given batting order (with a particular pitcher on the
mound) has of beating another lineup and their pitcher. This
probability implies what the betting line for that lineup should be
on the given day. If our model's suggested betting line does not
agree with the betting line available, the method suggests betting
on one team or the other. On the other hand, if our probability lies
within the spread, we suggest not betting on that game.
Various games are available at the
here for above average betting ability.
We include below a table relating betting lines to probabilities.
Betting Line 
Probability 
Betting Line 
Probability 
300 
75% 
+300 
25% 
250 
71% 
+250 
29% 
200 
67% 
+200 
33% 
180 
64% 
+180 
36% 
170 
63% 
+170 
37% 
160 
62% 
+160 
39% 
150 
60% 
+150 
40% 
140 
58% 
+140 
42% 
130 
57% 
+130 
43% 
125 
56% 
+125 
44% 
120 
55% 
+120 
45% 
115 
53% 
+115 
47% 
110 
52% 
+110 
48% 
105 
51% 
+105 
49% 
100 
50% 
+100 
50% 
Changes for the 2003 season
In August 2002, we found a silly coding error that meant all the
results from earlier in 2002 and all of those for 2001 were in
error. In the offseason, we reran the whole season after fixing
the bug (Aug. 20 until the end of the season were correct). For the
season, the method would have been up 640 units  using the 6%
cutoff described in the explanation for the 2002 season.
We thought that rather than using a set percentage error factor,
it might make more sense to bet only on games for which the expected
value of the winnings is greater than a given value. We performed an
analysis of the 2002 season results and came up with the table
below.
Expected Value 
Games 
Net winnings 
Win per bet 
0 
1457 
3357 
2.30 
4 
1132 
2822 
2.49 
8 
858 
3955 
4.61 
10 
724 
1165 
1.61 
12 
600 
1355 
2.26 
14 
522 
50 
0.10 
16 
424 
1450 
3.42 
17 
391 
2655 
6.79 
18 
359 
3395 
9.46 
20 
300 
2895 
9.65 
24 
200 
1305 
6.53 
30 
94 
60 
0.64 
35 
49 
890 
18.16 
40 
25 
1450 
58.00 
It seems that a reasonable balance between wanting to recommend a
goodly number of gains while maintaining positive results would be
only to recommend those games for which we have an expectation of
gaining 17 or more units. We toyed with the idea of recommending
games with regions of expectations (e.g. 18 to 30 and greater than
35) but since we have no explanation for why the results should be
better for 1830 but not slightly above, and the output gets more
complicated, we decided to keep things simple. (It turns out that
there is a negative expectation region for which the results were
quite positive  but, of course, we cannot explain that either). We
have used this expectation for the 2003 and 2004 seasons and plan to
use it for the 2005 season.
Changes for the 2002 season
In 2001, we recommended betting on a team if
the payoff (even if our model had an error of 2%) was favorable. An
analysis of that season's results indicated that we would have done
better (and much better per game) had we used an error allowance of
6%. We would have been ahead by more, while betting on less than
half as many games. We expect to recommend bets on far fewer games,
but hope the results will be better per bet.
Analysis of 2001 results:
See the table below for the results of the
analysis. The percent column gives how much error we allow our
method to have before deciding if the bet is favorable.
E.g., the bet is favorable if our model says the team
has a 60% probability of winning, we have a 10% margin of error and
the payoff is +100 or better  bookie says probability is less than
50%. That is, we bet when our probability is that percent or more above
the bookie probability based on the payoff. (In the analysis, we did
not consider games with no data, low data, pitching changes and
"over the limit" games).
Based on the table below, we decided that a buffer
of 6% allows for us to recommend plenty of games with a reasonable
amount of winnings per bet. Notice, that the numbers of wins/number
of losses is around 1, while the winnings were positive. That is
because the model tends to recommend more underdogs than favorites.
Percent 
Games 
Wins/Losses 
Total Winnings 
Winnings per bet 
0 
1444 
727/717 
3170 
2.20 
1 
1279 
648/631 
4785 
3.74 
2 
1141 
572/569 
3065 
2.69 
3 
1002 
502/500 
2700 
2.69 
4 
867 
441/426 
3545 
4.09 
5 
752 
380/372 
2825 
3.76 
6 
640 
332/308 
4640 
7.25 
7 
546 
284/262 
4160 
7.62 
8 
455 
234/221 
2495 
5.48 
9 
386 
195/191 
1290 
3.34 
10 
316 
155/161 
320 
1.01 
12 
211 
105/106 
765 
3.62 
14 
124 
63/61 
795 
6.41 
16 
79 
40/39 
830 
10.51 
18 
48 
28/20 
1440 
30.00 
20 
25 
14/11 
570 
22.80 
