OneSixThree Analytics @StatsIsKey

@CRichh12 Bc OPS+ is normalized across the league, dopers would spike up in relation to non-dopers. If testing then resumes, their OPS+ would fall back to their true averages. Not sure any model could measure of this, but if I knew a guy was doping and his FD cost was the same, Id pick him

@CRichh12 While I expect some sort of testing to resume once the lockout ends, this is still a very interesting question. Assuming there are some hitters who will be eager to cheat, I'd expect at least their SLUG to increase.

Hit a lull in posting but the work goes on. Currently researching machine learning methods to try to increase predictive power, but some infrastructure needs to be built to get ready for the next season (if we ever get one)

I've been taking a break for a while over the holidays but I'm back now and will resume updates on my progress. My first two tasks are to finish responding to the submitted questions and begin looking at ways to improve model performance at the beginning of the year

@BrewMke I don't have anything in the model right now, but I'm intrigued by the possibilities. So to answer your question, I'm going to create new variables that are moving averages of fanduel points with various lookback periods. I'll make a post with results when I have them.

I’d like to start making this account a little more interactive. Now that I’ve got a model, I’d love to hear from my followers what they think would be important predictors of daily fantasy points. So shoot me your ideas and I’ll tell you how/if that stat plays into my estimates!

@BrewMke For added application, a lineup with 8 hitters 1-4 in the batting order might average 120 points. While a lineup with equal cost using hitters 5-9 would average 108 points. This would take your expected win rate from 60% to 50%.

@BrewMke This graph shows expected change in points as you get further down in the batting lineup. The red line represents statistical significance, so we can say with confidence that 5-9 in the batting order will average less Fanduel points.

3. Each run follows the same general pattern (i.e. from days 60-90 every run goes bananas, right before day 150 every run drops, etc.) I'm not sure if this should be surprising or not. Maybe if I look more at days which are highly correlated across runs I could improve the model

2. The average max drawdown is $50 and the average finishing balance is $350. Say we start the year with $200 in the account just to be safe, we would estimate finishing with $550 (a 175% increase!)

Here are 5 runs using my current model to predict lineups in 2021 and pitting them against random lineups (again, assuming my opponent doesn't have a model themselves). This also assumes that I am playing 50, $1 games per day of the MLB season.

The current goal at this point is to use 2020 data to build a model and use it to predict lineups for each day in the 2021 season by the end of December

Progress is slow these days as I'm finishing finals, but hey, if the MLB is in a lockout I can take a break too

Going back through old code and updating has allowed me to solve the doubleheader issue that I had before. This will give me access to even more games in the year.

This time, with the knowledge I've gained I'll be updating my code to use the pandas library, which should speed up the process a lot.

Reflecting over the holidays has got me thinking that I need more data. I'd like to fit my model using a couple years of prior data and then predict the full 2021 season. This means its back to python and beautifulsoup to get more.

Thankfully some kind soul has this in excel format out there, so that will save some time. I'm very interested to see how this changes my existing model

Reading some analysis written by other fantasy gurus and they bring up using vegas odds as predictors, and I can't believe I didn't think of it earlier. Looks like I've got more web scraping to do.