The Coal Trader


Systematic Trading

Notes on developing a mean reversion systematic trading system.

Last updated as of: 8/20/2023


We’ve developed a short term mean reversion scalping trading system which is being tested live. Subscribers in the private Twitter feed can follow along with the trades and decide if they want to follow them or not. So far the results look pretty good, see below:

Since early June when this system was deployed into the real world, broad equity markets have been relentlessly trending higher. It’s therefore interesting how the short scalps worked well and the long scalps did not work so well. If we plug these results into the Kelly Criteria, we can see if Kelly himself would’ve recommend this system and perhaps his statistical method will even give us some trade sizing suggestions.

Kelly Criteria:

You can see the trade statistics in the above table. There have only been nine long trades so far, so not yet statistically significantly. I think we can therefore ignore the first column for now. We need approximately thirty trades to be statistically significant, and combining the Long and Short trades gets us above to that figure at thirty-three. Looking at the All column, you can see that this system has a 77% hit rate so far. The average losses are twice as large as the average wins however, which is not ideal. Nevertheless, Kelly suggests using 81% of capital for each trade. I obviously don’t use that recommendation verbatim, but I do look at it directionally in order to gauge a trading system’s performance and judge/compare them to one another.

The 81% figure is pretty good. My personal trading statistics since publishing and tracking them on The Coal Trader have the following Kelly scores: 74% for Longs, 14% for Shorts, and 32% for All combined. My portfolio is up around 500% in that two year period having those associated Kelly scores in approximately 1,150 independent trades. This suggests that if the statistics for this scalp system continue to hold true to form going forward, the two year performance could be in the range of 2x or 3x that of my own performance.

There are many caveats of course. For one, I have not been taking every trade signal which matches the conditions for the system. This is due simply to normal trading constraints and having too many signals, but also due to being too fearful at times. I’ve missed at least as many trades as you seen above. That being said, I think anecdotally the statistics would be even better if I were able to take every single trade signal. Another caveat is how I choose to execute each trade. As you can see I’ve only been using options. This works well for the quick trades, but has definitely worked against me on some that were a bit longer in duration; so option strike and expiration are additional variables to consider. Another caveat is that on a few occasions I’ve chosen to “double down” on a trade. I haven’t tracked this particular feature, but I can say that anecdotally it has both helped and hurt on occasion – probably creating a wash either way. 

Anyway, I will definitely continue to monitor the signals from my trading algo and attempt to take as many of the trades as I can. The statistics to date definitely appear to be something worth investing more time and money on. I plan on eventually publishing the trade signals on this website in real time, but there’s some programming that I need to work through first in order to pull that off. In the meantime, please make sure you’re following the private Twitter feed @thecoaltrader in order to see these trades in real time when I take them.