Algorithm and Portfolio Stats: 07/01/2024 - 07/05/2024
Our algorithm finishes its first week of the month almost completely in the black. To be clear, this indicates a pretty negative week. Overall, it’s looking like the experimental system we’ve been running is going to replace this one. The graph I have for the experimental system is a bit rough around the edges since it’s not optimized for a full release, but it should still give a good idea of how it’s performing.
While this one has its draw downs, they’re much smaller in both duration and magnitude than what we see from our main system. We only have 3 weeks of live performance from this algorithm, but through those 3 weeks and in backtests, we’ve seen much higher consistency of returns. At this time, the only thing I’m unhappy with on this system is the magnitude of returns. 0.2% in 3 weeks isn’t impressive, but if you’re cherry picking trades on this, it should be easier to increase that. In any case, increasing magnitude of returns is (on paper) the easy part - making them consistent is typically where the difficulty is, and that’s done to a satisfactory level here.
This algorithm still has further to go. As previously discussed, these systems are all vulnerable in that they rely on the TK-cross signal working. Since the TK-cross is inherently a trend-following signal, periods of serious market stagnation can drain profits here. As such, I still think consistency can be pushed to an even higher level by pairing this system with a mean-reverting strategy. This algorithm is currently making roughly 81 trades per day which - for the sake of the user experience - is a bit higher than we’d like. So, in the near future, it looks like our plan is to reduce the number of trades this system makes to see if we can increase profitability without significantly impacting consistency. After that, we’ll look into introducing a mean-reverting strategy once we have room for more notifications.
This system actually uses a new technique for us, which I also think will yield a stronger mean-reversion strategy. We’ve long been trying to introduce machine learning to our TK-cross systems, but their performance has always plateaued around ~2.3 Sharpe - about the same level of performance we see from our fixed rule sets. With this system, however, we have a new breakthrough. Instead of training just 1 model to determine winning/losing trades, we’ve trained 2 models: 1 that determines when to enter short positions, and 1 that determines when to enter long positions. From this model, we’re expecting a Sharpe of about 2.85 - a substantial jump in performance!
I intend to experiment a bit longer with the experimental algorithm, but it’s looking very likely that we switch it in as our main release in the near future.
All of that said, let’s examine the results from our portfolio this week.
Our portfolio absolutely ran away with it this week - beating SPY by 0.78%! Our bet on tech paid off this week, as the sector carried us to strong returns.
Naturally, we’re keeping largely the same exposure strategy going into this week. As always, our portfolio components for the week are listed below. For brevity, all tickers with allocations below 0.5% are excluded.
That’s all I have for you tonight. As always, thank you for reading and happy trading!
