My Arbitrage Algorithm Had a Great Backtest, But It Failed in The Real-World.
After digging into the data, I saw something scary. But I may just have a solution. Also, data storage, server infrastructure, and exchange agreements.
Background
About a month ago, I deployed my first multi-asset statistical arbitrage algorithm. I had a backtest sample spanning 25,000+ trades across 10 years, you can read up on it here.
The first few trades went fine and I was getting optimistic. However, as it traded more and more, the PnL started to go down; in essentially a straight line, despite the actual strategy not changing at all. So, after letting it go on to lose for another week, I dug into the data and found the culprit that revealed just how competitive the algorithmic trading space is.