carter

[carter @carterjcameron]

@quarccmtl Me when I'm long dave

[carter @carterjcameron]

Same lesson I've been applying since: build the pipeline so the research is repeatable. It closes the feedback loop and lets you improve faster.

Full writeup: https://t.co/yPirSouQVO

[carter @carterjcameron]

Back in January I competed in McHacks 13 with Samuel Beaudoin, Shuya Liu and Dan Moraru, and we won 1st place in the National Bank of Canada Capital Markets HFT Challenge.

I recently revisited our submission and wrote it up. A few reflections 🧵

[carter @carterjcameron]

I'd also build the flash-crash stop-loss we planned instead of relying on inventory limits. We gave back a lot of profit in the mini crash and barely made it above water in the full one.

[carter @carterjcameron]

@EnvolDev Cannot conclude with n of 1

[carter @carterjcameron]

@jonahseguin No

[carter @carterjcameron]

If you're evaluating any strategy, including your own, the walk-forward table is the one to look for.

Full writeup, with every table:
https://t.co/GfcRB8RKW0

[carter @carterjcameron]

🧵

Most of my recent research has been strategies nobody can realistically run at home, so I wanted to try something simple.

On a 10-year backtest it had equity-like returns and a much smaller drawdown than SPY.

However, as you will see these results were contaminated.

[carter @carterjcameron]

Same strategy in both charts. The only difference is what I was allowed to know when making the choices.

A clean 10-year backtest, a passing parameter sweep and a passing permutation test still didn't hold up once I stopped letting it see the future.