Weekly Algorithm Review: 10/28/2023 to 11/03/2023
Performance Rankings
Base Algorithm: +3.35%
Long Term Portfolio: +3.33%
Overall Market: +3.18%
Variable Sector Neutral: +2.51%
Variable Market Neutral: +2.06%
Sector Neutral: +1.05%
Market Neutral: +0.77%
For it’s first week, this iteration of the algorithm finishes in the black. The portfolio did well this week, outperforming the market by 15 bps. It does even better when compared to an equal-weighted ETF, which we consider it comparable to. It was a tougher week for small market cap stocks, so our portfolio (which weights these more than the S&P 500 does) to outperform SPY is a good thing to see.
We weren’t going to draw any meaningful conclusions about this iteration of the algorithm from a single week anyways, and performing so close to the long term portfolio, that goes double. Regardless, there are a few tentative observations we can make.
First off, this algorithm isn’t shy about picking standout tickers. If you were watching its recommendations this week, you saw the same thing I did: INTU, INTU, INTU. It broke 10% allocation every day this week, and broke 20% twice. Our returns by ticker will back this up. At first glance, that might seem promising, or at least reminiscent of what we saw with NVDA earlier this year.
Here’s why I think it might be a red flag: INTU wasn’t one of the better performing tickers we saw this week. In fact, it only ranked #58 out of 99 - putting it on the bottom half of our portfolio. In the spring, our algorithm put major allocations into META, GOOGL, and NVDA - but those all ended up among our top tickers in the portfolio that quarter. Obviously it’s only 5 days worth of data - in my opinion, no cause for concern, but something worth noticing.
Initially, my concern here was that the algorithm was trend-chasing, but I’m not convinced. In terms of total intra-day return over the last year, INTU ranks #25 from the top. In terms of percent of days that had positive return in the last year, it’s #59 from the top. INTU has had a solid year, but so have a lot of the stocks in our portfolio - it’s often a reason they’re in there to begin with. In my opinion, INTU hasn’t done well enough to be this over-allocated solely off of a trend-chase, at least not relative to the rest of our portfolio. Again, it’s only a week of data so we can’t draw any meaningful conclusion here, but my current hunch is that the algorithm has allocated into INTU owing to a strong technicals profile, moreso than any trend-following. This will remain to be seen.
This iteration of the algorithm has performed well in long-term backtests on a sample from the S&P 500, as well as a small backtest of our new portfolio. However, trend-chasing is my main concern with it. It doesn’t include our newer method of counteracting trend-chasing, and it increases the weight we put on older days in its training data. My concern is that, if one ticker in our portfolio has had a particularly strong year, the algorithm could lock onto it and not let go. With more recent systems, the negative emphasis on older data meant that such a ticker would at least have to continue having good weeks (or at least, mostly good weeks) in order to maintain its emphasis.
This is my main concern with this iteration of the algorithm, and one of the key things I’ll be looking out for during the beta. Over-allocating into a few tickers isn’t necessarily a bad thing. Even if those tickers aren’t the best in the portfolio, it can be completely fine - as long as the algorithm has strong reason to believe that said tickers will make good returns. I want to see if it knows when to take its foot off the gas.
What’s In The Pipeline?
Work is ongoing on a new intraday algorithm. We want to finish a non-trivial beta of this system before putting our weight behind anything new, but we still believe that this is our most promising option.
Additionally, we have been working on a proprietary Fear & Greed meter. As opposed to others, which give only 1 value, we intend to give 3. Our goal is to have 3 separate Fear & Greed meters - one for the short term, medium term, and long term. These will include unique factors to better suite the market’s sentiment over that time frame. We’re hoping that, together, these 3 meters will give a more in-depth indication of market sentiment at any given moment. For example, we might see the short and term leaning heavily into greed, but the long term more on the side of fear. This could indicate that the market expects a recession, but not in the coming weeks. More updates will come as we have them.
Misc. Data For The Week
