Weekly Algorithm Review: 12/02/2023 to 12/08/2023
Performance Rankings
Variable Sector Neutral: +0.56%
Overall Market: +0.53%
Base Algorithm: +0.4%
Variable Market Neutral: +0.29%
Long Term Portfolio: +0.16%
Sector Neutral: +0.16%
Market Neutral: -0.03%
This was a tougher week for the portfolio. We typically expect it to out-perform at least one of the ETF’s we compare it to, this week we’re 0 for 2. We expect this because, despite its allocations being equal in all of its component tickers, we try to match the S&P’s sector distributions. Regardless, this isn’t something we can expect 100% of the time. The portfolio operates on a longer horizon than any of our algorithms; we’d need to see a lot of weeks like this before it became a red flag. Not great to see, but inevitable. In any case, our portfolio has performed well since we’ve swapped to a beta-1 system.
For the second week running, the algorithms have performed well. 24 bps isn’t a huge win, but as always, every win is great to see.
The main reason the algorithm has performed fairly closely to the portfolio is that it doesn’t make major changes to its holdings. Realistically, we’ll only see a major divergence in performance if the algorithm moves heavily into 1 stock or 1 sector. This isn’t an unlikely scenario (we saw it throughout the spring and early summer, as well as a few weeks ago with TSLA), just not one we’ve seen this week.
Looking at things on an individual ticker basis, there’s really only one outlier this week, CHTR. It was the portfolio’s worst performer overall, and the algorithm did not under-allocate into it.
Almost all of the losses we saw from CHTR happened on Tuesday, the stock’s worst day this week. Ultimately, it seems like this one was just bad luck. Regardless, we still out-performed this week, so no use dwelling on one bad trade.
What’s In The Pipeline?
The new intraday system had a fairly erratic week. Its cumulative returns maxed out at roughly 0.2%, but it currently stands at -0.07%. Though negative, it’s still very close to 0 - even for this system. We’re still operating under the hypothesis that, in the long run, it’ll approach that 3.0 Sharpe we see in backtests, and even in the case it under-performs relative to these projections, we’re still expecting it to be profitable. Our testing suggests that the parameters we’ve chosen are not over-fit, but we’re yet to find a set of reasonable parameters (i.e. not intentionally trying to make it perform poorly) that result in a net loss during backtests. This means that even if the model doesn’t perform on the level of its backtests, it should still be profitable given a large enough sample size.
Before starting up this week, we’re doing a scheduled reset on it. Once every month, we’ll be having our intraday model select which tickers it wants to trade that month, and purging any existing positions that it hasn’t yet exited. Our backtests account for this, meaning that in order to replicate backtest results, we want to make sure we keep to this schedule. This insures that the model is always trading tickers it feels strongly about, while also giving it time to develop positions on tickers it considers to be worth trading. Obviously users won’t see any of this yet - this is still in an internal beta only - but something to expect should it move to a full release.
Misc. Data For The Week
