I have a table of tennis players where I've worked out the probability of them winning their respective matches in three different ways, e.g. % previous 10 games won, % first serves in play etc etc. I've assigned a weighting to each of these probabilities to create a blended probability which I convert into predicted odds. I then compare these predicted odds to the actual odds that are available from bookmakers to derive the 'edge'. A positive edge means the bookmaker is offering better odds than they should and there is 'value' in placing a bet. There are a three potential scenarios:
I want to work out what the optimal combination of the three probabilities AND the edge threshold that generates the highest level of profit. I tried to use Excel Solver to do this but apparently it can't cope with IF statements. I've read in a few places that you can convert IF statements into complex formulas that get around this problem but I haven't been able find an example that's clear enough to follow. What are the potential solutions?
For reference here is a sample of the table:
+-----+---------------+---------------+---------------+---------------------+----------------+--------------+--------+-----------+-------------+
| PKG | Probability 1 | Probability 2 | Probability 3 | Blended probability | Predicted odds | Actuals odds | Edge | WIN/ LOSE | Overall P&L |
+-----+---------------+---------------+---------------+---------------------+----------------+--------------+--------+-----------+-------------+
| 1 | 32.4% | 28.6% | 74.0% | 43.4% | 2.31 | 2.81 | 7.86 | 1 | 0 |
| 2 | 36.2% | 34.1% | 61.0% | 42.8% | 2.34 | 4.47 | 20.39 | 0 | -100 |
| 3 | 55.2% | 58.2% | 42.0% | 52.4% | 1.91 | 4.13 | 28.19 | 1 | 413 |
| 4 | 60.7% | 64.7% | 67.0% | 64.2% | 1.56 | 1.44 | -5.17 | 0 | 0 |
| 5 | 33.7% | 40.1% | 75.0% | 48.6% | 2.06 | 2.32 | 5.57 | 1 | 0 |
| 6 | 48.7% | 47.8% | 77.0% | 56.8% | 1.76 | 3.41 | 27.46 | 0 | -100 |
| 7 | 55.2% | 64.1% | 79.0% | 65.9% | 1.52 | 2.38 | 23.87 | 1 | 238 |
| 8 | 45.6% | 34.9% | 30.0% | 36.6% | 2.73 | 3.70 | 9.63 | 0 | 0 |
| 9 | 56.9% | 63.1% | 49.0% | 57.0% | 1.75 | 3.00 | 23.64 | 1 | 300 |
| 10 | 38.6% | 37.6% | 25.0% | 34.1% | 2.93 | 3.25 | 3.32 | 0 | 0 |
| 11 | 56.4% | 56.4% | 63.0% | 58.4% | 1.71 | 2.74 | 21.90 | 1 | 274 |
| 12 | 71.3% | 78.1% | 38.0% | 64.1% | 1.56 | 3.66 | 36.75 | 0 | -100 |
| 13 | 61.4% | 67.5% | 72.0% | 67.0% | 1.49 | 1.23 | -14.50 | 1 | 0 |
| 14 | 66.4% | 73.0% | 76.0% | 71.9% | 1.39 | 1.42 | 1.46 | 0 | 0 |
| 15 | 56.1% | 64.8% | 21.0% | 49.1% | 2.04 | 1.35 | -25.19 | 1 | 0 |
| 16 | 61.5% | 53.4% | 57.0% | 56.9% | 1.76 | 1.37 | 21.03 | 0 | -100 |
| 17 | 32.4% | 26.2% | 45.0% | 33.7% | 2.97 | 1.45 | -35.14 | 1 | 0 |
| 18 | 69.5% | 62.5% | 43.0% | 58.8% | 1.70 | 1.77 | 2.19 | 0 | 0 |
| 19 | 65.0% | 65.3% | 45.0% | 59.1% | 1.69 | 3.66 | 31.81 | 1 | 366 |
| 20 | 73.1% | 81.9% | 45.0% | 68.2% | 1.47 | 1.12 | -21.03 | 0 | 0 |
| 21 | 52.2% | 51.4% | 25.0% | 43.7% | 2.29 | 2.59 | 5.14 | 1 | 0 |
| 22 | 30.9% | 38.2% | 27.0% | 32.6% | 3.06 | 2.26 | -11.61 | 0 | 0 |
| 23 | 61.6% | 43.1% | 46.0% | 49.5% | 2.02 | 2.82 | 14.11 | 1 | 0 |
| 24 | 58.6% | 73.4% | 34.0% | 57.1% | 1.75 | 4.21 | 33.38 | 0 | -100 |
| 25 | 64.4% | 72.8% | 61.0% | 66.7% | 1.50 | 2.24 | 21.99 | 1 | 224 |
+-----+---------------+---------------+---------------+---------------------+----------------+--------------+--------+-----------+-------------+
Here are the weightings:
+---------------+-----------+
| Blend | Weighting |
+---------------+-----------+
| Probability 1 | 30.0% |
| Probability 2 | 40.0% |
| Probability 3 | 30.0% |
+---------------+-----------+
Predicted odds are calculated using the following formula:
=1/Blend prob
The edge is calculated using the following formula:
=(100/Pred odds)-(100/Act odds)
The current edge threshold is 20.
The IF statement to calculate the Overall P&L is:
=IF(WIN/LOSE=1,IF(Edge>=Edge threshold,100*Act odds,0),IF(Edge>=Edge threshold,-100,0))
I set up Solver as follows:
So I have tested your sample data using Solver with the following set up:
The Variables range is C29:C32
(the three weightings and the threshold);
The constraints are:
C29:C31<=1
as they are probabilities;C32<=100
must set an upper bound for this variable so I picked 100
which should be safe?C33=1
which is the sum of the three weightings which should be 100% exact?Use Evolutionary method I have obtained the following:
2210
;37.5%
9.1%
53.4%
18
I think the WIN/LOSS result plays a significant role in this exercise but what would be the reasonable constraints for this range except
binary
?