I have an asinine idea and it's terrible.
So, as a flight of fancy.... Here's my shitty diatribe.
I'm bored shitless at work and I want to use statistics somewhere in my data daily trend software. So, while googling around and listening to cute Chinese girls talk about data science. I came across something called the One Armed Bandit and the Multi-armed Bandit.
Essentially, you can predict with a large degree of accuracy which slot machine will pay out the most based off a small number of win/losses. The problem is, slot machines most likely have a moving average of wins in order to combat the next statistician from coming and making bank. You can still use a Poisson's Distribution to determine when high pay out times are. At least I reckon.
The national and state lottery is obviously stacked against you and it would take a fair degree of the most pulled number combinations to actually win the MegaMillions or PowerBall.
Ergo, the rational conclusion is scratch offs. They have a fixed statistical win rate (Usually around 1in5 for $1 tickets and 1in3.58 for $50 tickets).
I pulled off this trick on a $1 scratch off and bought five $1 scratch and only won once. Therefore, my win rate is 1in5. The lotto website states that game has a 1in4.85 chance of winning. Not far off at all. I caught it at the right moment and it gave me the number stated on the website.
Where does multiarmed bandit come into play?
Let's say you visit several (about 4) gas stations and buy 20 $1 lotto tickets. That will be ~4 winning tickets out of 16 tickets. The ones with the biggest pay out are can therefore be calculated through Thompson Sampling and Beta Distribution and you can make a pretty safe bet that the roll the highest payouts came from is most likely going to continue that trend.