- Credits to n00n and Jasper for Open sourcing their backtesters/visualizers and working so hard to debug them.
This is work from someone learning about algorithmic trading as well as working on improving CS/Coding skills as I am learning more or less on my own (Try not to use R/R Markdown). Currently top 30 working alone out of 1700+ active teams and 7000+ total teams who dropped out at one point or another. Overall, I would first like to thank IMC Trading for the oppertunity for not only running an incredible competition but also designing it in a way that challenges what seems to be multiple asepects of a quant trader while also doing so on a smooth and good learning curve starting with a simple task to more and more complicated tasks.
Leasons:
- Slow down, try not to do too many things at once.
- A good backtester is just important if not more than coming up with strategies
- Keep note of specific changes you make to accomidate limitations of backtesters (this costed me around 70k + and probably top 20)
- Understanding what might be Game Theory optimal may be nice but understanding your opponents is much more important. I was supprised by the amount of people who voted on the second manual trading round, wasn't as self selective as I thought and missed profitability range by 2-3 hundred points. (exected loss of around 50K)
Strategies:
Now most of the main ideas were more or less obvious for each, but
-
Pearls
- Market making and arbitrage of 10k
-
Bananas
- Market making with a bit of directionality
-
Coconuts/Pina Coladas
- Pairs trading off of distribution of the difference and formulating a reward function to determine function of positional limits
-
Berries
- Time series based, hard coded most of this based on momentum and time thresholds
-
Picnic/Dip/Ukulele/Baguette
- Pairs Trading again but with more items where 3 are liquid but the last is very illiquid
- decided to stay as neutral as possible, may have costed me more potential profit
-
Gear/Dolphins
- Event/News based
- Deciphering what is and isn't a signal, which saved me from potential loss in Round 4
-
Counter Party
- Mainly tried finding potential positive strategies not market making (Caesar)
- Maybe should have tried some sort of ML signal rf but lacked the time
- Lots of "noice"/useless information and signal
- Mainly traded on Olivia who was basically a prophet and knew the daily highs and lows.
- Optimized Berries to include her info
- Took a risk on the Pairs Trading on Picnic Baskets with the high and low signals on Ukulele
- Since Picnic = 4 * Dips + 2 * Baguette + Ukulele + 360
- We know that Ukulele = Picnic - 4 * Dips - 2 * Baguette - 360
- Once given either signal go long one way and short the other
- Depending on the chart for Round 5 this may hurt or benefit me.