Algorithm and Portfolio Stats: 05/27/2024 - 05/31/2024
This week, I’m bringing an update on our algorithm project. Having spent most of this month heavily auditing our live results, we have a system that we feel good about releasing. Our chosen system does not utilize machine learning. After some heavy auditing of trades, and reworks of our live engine, we’ve seen its live trading exactly match our back tests of it.
On a back test since March 19th (the original date we rolled out our current algorithm), this new version has a Sharpe of 2.35. Notably, some of our other experimental systems have marginally higher Sharpes than this over the same test period. We’ve chosen this system because it hits “higher lows” than those comparable to it. To elaborate, April 4th was a great day to be trading TK-crosses. Of our 6 experimental systems, 4 of them use the TK-cross strategy. Of these 4, all of them had their best performance on that date - some by higher margins than others. All of our TK-cross systems have inferior performance when we isolate them from this day. Our chosen system has a substantially higher Sharpe than any of our others, when isolated this way. For this reason, we’re expecting more consistency looking forward.
I’d like to take a moment to recap what exactly we’ve changed here, as a result of our mass auditing. I’ve discussed these during the last few weekly reviews, but to recap:
Some of our live implementations of our algorithms had a bug where they would be stopped out prematurely. This has been fixed, and recent live records are now matching back test stop points correctly.
Our live implementations were using the entire S&P 500, whereas our back tests only used a subset that hit certain liquidity requirements. As such, our live algorithms under-performed in part because they were running on stocks less liquid than those they were trained on. This has also been corrected.
Our backtester had a bug where it wouldn’t account for a moving stop loss until 5 minutes after the trade had begun. The impacts of this were fairly negligible relative to our other changes, but I’m including it here for completeness.
So, with changes made and a new system selected, what comes next?
We’ll be running a more in-depth auditing process on our systems. So far, we’ve periodically reviewed stock charts of its trades to make sure they look attractive to a human trader, with a reasonable rate of exceptions. This will continue, but the problem is chart review doesn’t catch bugs like stop-outs 1 candle early, or the wrong stocks being traded entirely. From here on out, we’ll be incorporating a regular numerical audit, in which we manually comb through the parameters of each trade we see live, and compare them to the numbers we see in back tests.
Research on machine learning systems will continue. They’re currently on par with our best non-ML systems, but with “lower lows”. I feel that by continuing to move forward here, especially with our fixed live bots, we can reach new heights of consistency. For one thing, I’d like to segregate things into 2 models here. 1 for evaluating if a cross is worth trading, and 1 for individual re-entries.
Research will also continue on mean-reverting strategies to pair with our TK-cross system. These aren’t at the level we want to see yet, and are mostly going to take a back seat to TK-research, but this isn’t something we’re abandoning entirely.
Lastly, we’ll be rolling out our new algorithm on Monday. Expect a larger number of trades here (about 40 per day, compared to ~10 from our most recent system).
Now then, let’s examine our portfolio this week.
This was a tough week for our portfolio, under-performing SPY by a little over 0.6%. The main reason for this loss was the tech sector. It’s our biggest allocation, as is typical for our momentum-emphasizing strategy, but with the exception of AAPL and NVDA, we didn’t have any solid winners from it.
One bad week isn’t a red flag, especially given our recent performance, but regardless we’re proactively determining our future strategy. For the time being, we’re continuing our tech-heavy strategy, but should this trend continue in the markets, we’re prepared to pivot.
The contents of our portfolio for this week are listed below. As always, tickers with allocations below 0.5% are excluded for brevity.
That’s all I have for you tonight. As always, thank you for reading and happy trading!