This is another awesome post! Thank you for sharing your insights and expertise, am so grateful. I have had a go at recreating what you have shared, and I experimented with the parameters. I settled on 70% gap up and a SL of 30%, which gives a decent enough looking equity curve, (can't past chart here unfortunately), but not as good as yours. I think the SL is the problem. I tried 2 x ATR and fixed $ but still not as good as yours. Any pointers which will help me in the right direction? Thanks very much
Yes, it's way less profitable but there are many profitable large caps systems. For example there is one published by Laurens Bensdorp in his book The 30-minute stock trader. Also the one I wrote about here is mainly large caps - https://stonkscapital.substack.com/p/systemizing-kullamagis-parabolic
Hey this is more relating to the first part but i'll ask it here anyway. I am currently automating my small caps strategies in Sterling ActiveX API (cheaper than das and atm i can't afford DAS), and i am wondering how you deal with halts, because in case there was a 1000% halt like you said there once was, will your account simply get blown, or is there any way to prevent that from happening?
Thanks and much love for the work and alpha you share with the community!!
okey thanks, and i always wonder what data you're using, because i've heard lot of datas like polygon and stuff aren't good at all and i was wondering with your real market experience which one has been the best overall. Thanks!
Hey Aniol, question, which broker are you using with the Sterling API? For Cobra, I thought Sterling was more pricey that DAS. As a side note, I got DAS to email me their client API code, and was not impressed. And also I heard from someone on the inside of 1 of the biggest clearing firms that Sterling's API is much higher quality.
I use Zimtra, i am building it at the moment and in my opinion you can do everything you need (at least what i need because my bot is not for hyperscalping and stuff like that). In Zimtra you pay 60$ (if your buying power is less than 250k (less than around 75k balance) and 195$ for API. This prices are paid each 1st of month.
1. Divide the data into a few groups randomly, optimize based on small amount of parameters (I like up to 3 on 1000 of instances). Test if your expectancy is the same after applying the new parameters on out of sample data.
2. Keep it REALLY simple.
3. Use parameters that makes sense logically and try to come up with them on your own without randomly testing a ton of combinations until you find something that works
I have a list of delisted stocks from Polygon and some of them have earnings in FMP. Never really did a test of what percentage of the delisted db ACTUALLY have earnings reports in FMP, I need to check that.
This is another awesome post! Thank you for sharing your insights and expertise, am so grateful. I have had a go at recreating what you have shared, and I experimented with the parameters. I settled on 70% gap up and a SL of 30%, which gives a decent enough looking equity curve, (can't past chart here unfortunately), but not as good as yours. I think the SL is the problem. I tried 2 x ATR and fixed $ but still not as good as yours. Any pointers which will help me in the right direction? Thanks very much
can't share stops but just keep experimenting
Sometimes, small caps are difficult to short. Did you try the approach with eg. S&P 500 stocks? Would be very interesting to hear your thoughts.
Yes, it's way less profitable but there are many profitable large caps systems. For example there is one published by Laurens Bensdorp in his book The 30-minute stock trader. Also the one I wrote about here is mainly large caps - https://stonkscapital.substack.com/p/systemizing-kullamagis-parabolic
Hey this is more relating to the first part but i'll ask it here anyway. I am currently automating my small caps strategies in Sterling ActiveX API (cheaper than das and atm i can't afford DAS), and i am wondering how you deal with halts, because in case there was a 1000% halt like you said there once was, will your account simply get blown, or is there any way to prevent that from happening?
Thanks and much love for the work and alpha you share with the community!!
There's no way to protect against the 1000% halt, not with automation or manual. You can only protect yourself with size.
okey thanks, and i always wonder what data you're using, because i've heard lot of datas like polygon and stuff aren't good at all and i was wondering with your real market experience which one has been the best overall. Thanks!
Polygon is good enough and match my live trading
alr thanks man!
Hey Aniol, question, which broker are you using with the Sterling API? For Cobra, I thought Sterling was more pricey that DAS. As a side note, I got DAS to email me their client API code, and was not impressed. And also I heard from someone on the inside of 1 of the biggest clearing firms that Sterling's API is much higher quality.
I use Zimtra, i am building it at the moment and in my opinion you can do everything you need (at least what i need because my bot is not for hyperscalping and stuff like that). In Zimtra you pay 60$ (if your buying power is less than 250k (less than around 75k balance) and 195$ for API. This prices are paid each 1st of month.
How do you manage overfitting? Thanks!
1. Divide the data into a few groups randomly, optimize based on small amount of parameters (I like up to 3 on 1000 of instances). Test if your expectancy is the same after applying the new parameters on out of sample data.
2. Keep it REALLY simple.
3. Use parameters that makes sense logically and try to come up with them on your own without randomly testing a ton of combinations until you find something that works
Thanks for the reply! You're in sample is usually 80% of the dataseta?
more like 50% on small caps
Just to let you know, it's your fault if I'm starting shorting small caps! 😄
Have fun with the infinite risk !
Really helpful post thanks for sharing!
Do you have any pointers as to what overall profit factor to expect so I know I'm on the right track?
Hi interesting! How do you get from FMP historical data from delisted stocks? So that you avoid survivorship bias in the backtest?
I have a list of delisted stocks from Polygon and some of them have earnings in FMP. Never really did a test of what percentage of the delisted db ACTUALLY have earnings reports in FMP, I need to check that.
I am interested—how many stocks do you usually use for backtesting your strategy? And how far back do you go?
I go as far back as Polygon let's me - 2008. Some will say going back to say 2017 is totally enough
And do you have any rules? For example, I want to test a strategy on at least 50 stocks. How do you decide which stocks to use for testing?
Not sure what you mean. Feel free to DM
Hey Niv, do you still do your backtesting with Python or do you nowadays use Spikeet?
Python