A.I.M. Users Bulletin Board

[JoeForkeyBolo]

I believe I posted how I selected starting equity when I first posted my backtesting results from Q1.

1. What made you decide on the starting equity levels?

In selecting the starting equity level, I typically reviewed at the preceding 1-2 year stock chart and evaluated where the starting price was in relation to that price range. If low, I generally chose a higher starting equity >50%. However, as you can see I started with only 25% for HIBL, a new addition to the 3x S&P 500 Large Cap Index ETFs. I backtested HIBL starting at inception date. The only prior history I had was other 3x-leveraged S&P 500 ETFs. For HIBL, the other ETFs were near the highs, so I elected to start that one at 25%.

In hindsight, I should have started at 50%, but I think the knowledge of the Covid-19 impact on the 2020 market might have influenced my decision. I will likely re-do the HIBL backtest for 50% equity to be a fairer comparison to the other S&P 500 ETF results. You will notice no other ETFs have start equity below 50%. Now that I have the drawdown curve strategy in place, I don't fear a downturn will result in running out of cash, so I don't try to preserve higher cash levels at the initial investment, even if the price is at or near a 2-year high.

2. What settings have you been using in AIM

Settings for Buy SAFE • Sell SAFE • Minimum Transaction I've used in backtesting have been:

10% • 10% • 10%          10% • 10% • 5%          7.5% • 7.5% • 5%          5% • 5% • 5%

For funds I've tested with multiple settings, I've only posted those with the best overall returns. The settings chosen depend on the volatility (standard deviation) of the ETF. Those with higher volatilities get the 10 • 10 • 10 treatment. The lowest volatilities get 5 • 5 • 5 settings. I'm still working through testing for crossover points and plan to eliminate one of the two middle settings. I had worked with the Grok3 AI model to determine crossover points, and posted results here, but the AI crossover points were quite a bit higher than those in my backtesting. I believe the difference is due to assumptions made by the AI model that were not aligned with my method.

I have been working on C++ code that will read my historical price file, implement the preemptive AIM model the same way I do it manually, and output results for the various settings. This would be much faster than manual testing using the various settings. I've taken a break from all the backtesting and crossover testing to focus on other things.