Inspiration

I liked the stock prediction track, since it had "pure" results in so much that you could tell immediately where you were relative to others.

Further, in so much that the efficient market hypothesis holds, it should be very difficult to get anything working, especially since we seemingly have a bunch of "low signal" covariates.

Additionally, in my opinion, it was the best "solo" project, given my extant circumstances.

What it does

Offline, we fit a model. "Live," it ingests previous day's data, apply pertinent transformations, and then makes a prediction for that current day's trading.

How I built it

This was using jupyter notebook on my local MBP, which my MBP wasn't super forgiving, since it was pegged at 100 cpu a lot.

Challenges I ran into

Wife works on the weekend, so I watch the three kids, which means my "work time" was during my youngests' nap time (the other two are reading or watching PBS kids) and after I put them down.

There was also an issue with the simulator/submission process, which forced me to go to bed at 12:30 AM. This was a blessing in disguise, since I really needed the rest as well as my kids don't want to see (too much of) a sleep-deprived father.

Technically, challenges were that the initial approach used deep learning methodology but given that there were ~900 data points, I had to scrap that approach and choose something else.

Further, this wasn't a simple prediction problem, i.e., minimize RMSE. There's a notion of trade-tactics, i.e., how much you want to buy/sell, that makes this a joint optimization problem that's not convex.

Accomplishments that I'm proud of

Personally, going to bed at a reasonable time (and consequently making breakfast for the kids with a good attitude).

Even though I couldn't join a team (shoutout to my cohort at Texas A&M MS Stats people) due to my weird schedule, I was happy that I was even able to submit something.

What I learned

Sometimes, you need to take a step back and assess the situation of the data and the approach. That is to say, sometimes you're unknowingly stuck at a local minima and you have to crank up the \eta to get out of there until you get a better lay for the land.

What's next for Lone Wolf and Cub(s) Stock Prediction

Teach my cubs to intelligently mash buttons in a jupyter notebook, while I sit back and supervise.

Built With

Share this project:

Updates