When I was a wee lad playing a card game the family called No Name, I said what has become one of my most quotable quotes within the family. Feel free to let me know what the game is actually called, but basically I think it's like an individual euchre where you bid on how many tricks you think you'll win. Overbidding lost you points, while underbidding left points on the table. I suppose I must have been a pretty confident child so I made what may have seemed like some pretty outrageuos bids. So halfway through the game when my dad thinks I'm making a big mistake by bidding so high, he must have said something along the lines of "let me explain to you how this works, kid." Knowing where I stood on the scorecard and where he stood, I responded as sassily as I could, "You teach me?"

So following what has been called by at least one sports commentator the biggest upset in sports history, I feel the need to educate everybody who sees how terribly my algorithm is doing on the leaderboard and thinks "stupid Brandon, thought he could predict a perfect March Madness bracket." First of all, read the Understanding the madness section in one of last year's March Madness posts to see that predicting a good bracket is not the objective of my algorithm. No matter how many times I tell people this, I always get the question, "how's your bracket doing?" with a smirk on the face of the person asking. To which I typically respond, "not well, but I couldn't care less about my bracket." I root against my bracket approximately 50% of the time.

Now back to the biggest upset at least in the history of March Madness. Only halfway through the competition and I can tell you that I would have performed better had I simply submitted 50% probabilities for each game. Even if I predicted the rest of the games in such a way as to receive no penalty. This is because I employed a common strategy where you predict a probability of 100% for 1-seeds to beat 16-seeds. First of all, this is a strategy that bit me in the butt. It's not the result of a poorly optimized algorithm. So that begs the question, "why not just use the non-poorly optimized algorithm so you don't get bit in the butt?" Good question. That is what I want to address here.

Below I have some simulations you can look at or play with in the demo section. But here's the explanation. The simulation will be in terms of March Madness, but I'm going to explain it in terms of something I know much less about, but the average reader might know more about. Let's pretend we have a giant game of poker (perhaps online) with 1,000 people participating (I imagine that's impossible for most variants of poker but just pretend it's possible). Everybody is playing for money and has no knowledge of the probabilities of certain events. However, let's say there's an in depth hint section that evaluates the situation and can tell you all the probabilities and how you should expect them to impact you financially. Nobody knows it but this program is actually only giving the true probabilities to one of the thousand players. Everybody else is getting some random deviation from the true probabilities. Now the question is, who is most likely to win?

Basically, the takeaway I want to note from these live simulated graphs is that those receiving deviated probabilities can actually have a greater chance of finishing in first place. The catch is that they also have a greater chance of finishing in last as well. In other words, use the real probabilities if you care about finishing in a respectable position. If you care about beating the rest of the competition, take risks. If you care about both, take risks, then write a blog post explaining why you did what you did so your friends and family don't think you're an idiot. Deviating from the model's predicted probabilities for the 1 vs 16 -seed games is an easy modification that would have only helped people in years past. One of the so-called experts in the competition who is suffering even worse than I am from the risk of deviation from the model in the Men's tourney comeptition is actually winning in the Women's.

Lastly, just want to note that I like to consider myself a fan first. So that upset that killed my algorithm was awesome. I started participating in the algorithm competition because I enjoyed watching March Madness - not the other way around. So wipe those smirks off your faces next time you ask how I'm doing. :P

If you want to play with the graphs in a demo, you can check that out here.