Weekly Algorithm Review: 12/11/2022
Algorithm Performance This Week
Market Neutral: +1.63%
Variable Market Neutral: +0.53%
Sector Neutral: -0.25%
Base Algorithm: -0.45%
Variable Sector Neutral: -0.58%
Long Term Portfolio: -1.13%
Overall Market: -2.55%
I’ll be up front, this was a great week for us. Almost everything did as we expect. The market was generally down, and our defensive long term portfolio outperformed it. Aside from right after open on Monday, our portfolio was ahead of the market for the entire week, never falling behind it.
Our main algorithm outperformed our portfolio as well, giving us another leg-up on the market. Our market neutral and variable market-neutral systems outperformed everything else we had, which is normal when the market as a whole has a red week like this. With MN’s having near-zero market beta, it was able to gain a strong positive return, beating the market by more than 4% this week!
Our only weak spot here is our sector neutral systems. It might seem counterintuitive that hedging against market sectors would do so little for our returns, so I’d like to shed some light on why this is.
Many of the sectors our system is most exposed to (Consumer Staples, Healthcare, Utilities) either did well, or saw minimal downside this week. For this reason, hedging those exposures away didn’t help the sector neutral systems much this week - and in a few cases actively worked against it.
Variable Sector Neutral was also rather interesting. Normally, we expect it to spend most of the time between the base algorithm and sector neutral, but this week it finished outside of that range. Though an unusual pattern, I don’t think this is cause for alarm at this time. It looks like the only reason this happened was the base algorithm and sector neutral finishing so close together, coupled with the normal variance of trading. We can see that it mostly stayed between them, and even more so the previous week. Should this pattern become a regular occurrence, it’ll be something to look into.
Future Developments - What’s In The Pipeline?
We have spent the last week working mainly on the “hard rules” algorithm mentioned in last week’s review. We’re going with a genetic algorithm to train it, which is currently operational.
We’re doing some initial testing with it before going for a long-term training session. Our goals are to find out the following:
Under our current design of “trading rules”, does a genetic algorithm improve at trading its testing data set over subsequent generations?
Under that same design, do improvements in the testing data carry over into validation data?
Does this system of learning work better with multiple tickers simultaneously, or by trading 1 ticker at a time?
If these results are promising, we could be rolling out a new system in the next few weeks. If they aren’t, we can always develop the system further and try again. I like to test out basic versions of ideas first, because I’ve noticed we have a tendency to overcomplicate things to the point of diminishing returns. But with any luck, we’ll have an update next week on how this has gone.
That aside, I’d like to roll out different weightings for each user in the market game. When we started out this month, the algorithms just ignored the first 5 submissions from each user, when determining how aggressively to hedge. But as more user data come in, we can roll out more complex systems for this. This is something I can’t be too specific on (don’t want people gaming it), but it’ll do us all a favor in the long run. I don’t know about you, but I don’t want an algorithm controlled by someone that we know doesn’t do well.
Extra Data
