Testing the “all-in” strategy with data

Does the “all-in” strategy actually work? CPP reader Edgar Medina (@Ironclad) ran some computer simulations with the probabilities in the table from week 9, and tested the money picks (the blue, yellow, and green shaded picks). He tests the Giants (green money pick) at various confidence levels, all the upset picks, and various other scenarios using the two leagues he’s part of. Here’s his write up of the work he did. Thank you Edgar! 

—————

Kickoff time is here!  It’s too late to change our picks, but we can study what might have been.  The retrospective simulations shown here are for a 32-player league, which I think is on the small side of how people play.  They are retrospective because they use actual pick data from a specific league, looking at how different pick sets would have performed.  The win probabilities from the Week 9 Update table are used to randomly draw the winners of the football games for each simulation.  I also have a 77-player league, so let me know if you would like to see the results from that larger, more competitive league.

We start by picking straight-up by the win probabilities. Note that in my league the confidence points are scored from 16 to 4 instead of from 13 to 1:

SEA 16, CIN 15, SF 14, KC 13, CLE 12, IND 11, DEN 10, NO 9, DAL 8, MIN 7, PHI 6, MIA 5, BAL 4

Here’s what we can expect from this pick set in this league:

Simulating 400000 draws...
Average Points: 91.2
Median Points: 92
Average Finish: 13.4
Median Finish: 13
Chance of 1st Place Finish: 2.7%
Chance of Top 2 Finish: 9.6%
Chance of Top 3 Finish: 14.0%

Now let’s look at Dale’s top upset picks. We’ll start by looking at NYG, which has a 38% chance of beating IND.

% Win % Top 3 %
% IND 11 2.2 9.1
% NYG 4 8.6 17.3
% NYG 5 9.0 18.2
% NYG 6 9.6 18.8
% NYG 7 9.9 19.6
% NYG 8 11.9 20.9
% NYG 9 12.7 21.8
% NYG 10 12.7 22.8
% NYG 11 14.0 24.1
% NYG 12 14.2 24.4
% NYG 13 14.8 25.1
% NYG 14 15.2 25.2
% NYG 15 15.5 25.1
% NYG 16 15.6 25.4

(These results are generated by running 40,000 random draws, which tends to give results that are within half a percent. This accounts for a little bit of noise in the results, but is quick and good enough, I think. For example, you can tell from looking at the table that if I ran NYG 10 again it might come out around 13% Win%.)

My conclusion: If you’re going to pick only one upset and you’re going for the weekly win, by all means go ALL IN with MAX points!

Now let’s look at the other upset pick, HOU over PHI. By itself, we know that if we’re going to pick it, we’re going to max it out. (When I run just one case instead of a sweep like above, I run 400,000 draws to get a more accurate answer.)

=== HOU 16 ===
Simulating 400000 draws...
Average Points: 87.8
Median Points: 88
Average Finish: 17.3
Median Finish: 22
Chance of 1st Place Finish: 14.4%
Chance of Top 2 Finish: 22.2%
Chance of Top 3 Finish: 26.3%

That’s another pretty good pick! Each of those upset picks by itself gives us >25% chance of being in the money (if “in the money” is Top 3) Now what if we combine them? We have a 17.5% (38% x 46%) chance of hitting both these upset picks.

First, with minimal points (4 & 5). So our picks are SEA 16, CIN 15, SF 14, KC 13, CLE 12, DEN 11, NO 10, DAL 9, MIN 8, MIA 7, BAL 6, NYG 5, HOU 4:

=== NYG 5, HOU 4 ===
Simulating 400000 draws...
Average Points: 89.2
Median Points: 90
Average Finish: 16.2
Median Finish: 17
Chance of 1st Place Finish: 9.8%
Chance of Top 2 Finish: 15.2%
Chance of Top 3 Finish: 18.8%

Not bad, but not as good as maxing out either single upset pick. Now, what if we max them both out?

=== NYG 16, HOU 15 ===
Simulating 400000 draws...
Average Points: 83.4
Median Points: 84
Average Finish: 19.9
Median Finish: 23
Chance of 1st Place Finish: 10.6%
Chance of Top 2 Finish: 14.9%
Chance of Top 3 Finish: 17.2%

Interesting. We have a slightly higher chance of winning if we max out these upset picks (compared to minimal point upset picks), but the chances of being in the money are slightly lower, and your overall expected finish suffers, so if you’re hoping to stay in the running for season winnings, don’t do this!

Now I want to make sure that NYG and HOU are in fact the best upset picks. Let’s max out each potential upset to 16 pts, one at a time, to make sure that we’re not leaving anything on the table:

UPSET PICK WIN % TOP 3 %
OAK over SEA: 7.1% 9.9%
JAC over CIN: 11.3% 15.3%
STL over SF: 11.2% 16.5%
NYJ over KC: 11.8% 17.4%
TB over CLE: 14.6% 21.7%
AZ over DAL: 7.3% 13.2%
NYG over IND: 16.0% 25.7%
CAR over NOR: 12.1% 23.5%
NE over DEN: 11.8% 24.2%
SD over MIA: 3.5% 13.6%
HOU over PHI: 14.6% 26.3%
WAS over MIN: 9.5% 26.7%
PIT over BAL: 4.3% 15.6%

Yes, the pick table is correct! NYG and HOU are the best upset picks. TB over CLE is interesting as well, and arguably should have some shading in the table.

Conclusion: Pick one upset and max out the points. Do not pick two upsets.

Now let’s look at favorites. I will state without showing the data that the pattern on bumping up favorites is the same as for upsets: If you’re going to pick just one, you want to max out the points! Sweeping over the favorites, bumping each individual one up to the max of 16 pts:

FAV PICK WIN % TOP 3
SEA over OAK: 2.6% 13.6%
CIN over JAC: 2.7% 13.9%
SF over STL: 3.3% 13.6%
KC over NYJ: 3.4% 13.7%
CLE over TB: 3.4% 12.3%
DAL over AZ: 6.2% 18.2%
IND over NYG: 4.2% 11.6%
NOR over CAR: 6.1% 13.5%
DEN over NE: 7.9% 13.3%
MIA over SD: 8.3% 22.0%
PHI over HOU: 4.4% 11.8%
MIN over WAS: 5.3% 15.1%
BAL over PIT: 10.0% 23.8%

The most lucrative favorite max bumps are the ones that are shaded in the Week 9 Update table. DEN over NE is also somewhat interesting.

Conclusion: In general, picking any one favorite at max points is not as lucrative as picking one good upset at max points.

But…what if we combine some favorites with some upsets?

HOU + DAL, both max pts:
=== HOU 16, DAL 15 ===
Simulating 40000 draws...
Average Points: 87.1
Median Points: 88
Average Finish: 17.5
Median Finish: 22
Chance of 1st Place Finish: 14.3%
Chance of Top 2 Finish: 21.4%
Chance of Top 3 Finish: 25.9%

DAL max pts, HOU min pts:
=== DAL 16, HOU 4 ===
Simulating 40000 draws...
Average Points: 88.4
Median Points: 89
Average Finish: 16.9
Median Finish: 18
Chance of 1st Place Finish: 9.8%
Chance of Top 2 Finish: 16.2%
Chance of Top 3 Finish: 20.4%

And some other max-max combinations:
=== NYG 16, DAL 15 ===
Simulating 40000 draws...
Average Points: 85.8
Median Points: 86
Average Finish: 18.7
Median Finish: 24
Chance of 1st Place Finish: 15.1%
Chance of Top 2 Finish: 20.5%
Chance of Top 3 Finish: 25.0%

=== NYG 16, MIN 15 ===
Simulating 40000 draws…
Average Points: 85.0
Median Points: 86
Average Finish: 19.3
Median Finish: 24
Chance of 1st Place Finish: 13.6%
Chance of Top 2 Finish: 17.5%
Chance of Top 3 Finish: 20.3%

=== NYG 16, MIA 15 ===
Simulating 40000 draws…
Average Points: 85.5
Median Points: 86
Average Finish: 18.9
Median Finish: 23
Chance of 1st Place Finish: 15.5%
Chance of Top 2 Finish: 19.1%
Chance of Top 3 Finish: 21.1%

=== NYG 16, BAL 15 ===
Simulating 40000 draws…
Average Points: 84.9
Median Points: 85
Average Finish: 19.3
Median Finish: 23
Chance of 1st Place Finish: 13.9%
Chance of Top 2 Finish: 17.3%
Chance of Top 3 Finish: 19.2%

There’s definitely some good fav-dog combos in there, but none are appreciably better than picking a single upset at max points.

I definitely want to look at a larger league size to see if these results hold, but for now it’s time to watch some football! My picks for the week: HOU at 16 + DAL at 15. Go Texans! (And I can’t believe I’m also rooting for the Cowboys!)

  • Kyle

    Thanks for doing this! Great stuff, would love to gain some insight on how you created the table, I’d be interested in trying to replicate it.

    • Hi @disqus_gG2aPQAREv:disqus , this was done by @Ironclad with some simulation programing with data from his personal league. If you do it I’d love to see what conclusions you come to!