Fully Automated Stocks Trading
- Thread starter ValeryN
- Start date
- Status
I did at some extent and it is not something that appeals to me. Mostly because
(a) it tends to use quite a bit of capital
(b) works until it suddenly doesn't
(c) returns are somewhat low
And using leverage to improve returns seems too risky because of b.
It will be very interesting to hear from individual traders with solid experience in that area though.
Thanks for the solid replies.
Sorry for all these questions but I am really passionate about trading and always willing to learn from others doing similar stuff.
Would be nice if you could talk about how you optimize capital allocation throughout your different strategies? Are you optimizing to minimize specific metrics? Are you using more of a common sense approach?
Thank you!
Sorry for all these questions but I am really passionate about trading and always willing to learn from others doing similar stuff.
Would be nice if you could talk about how you optimize capital allocation throughout your different strategies? Are you optimizing to minimize specific metrics? Are you using more of a common sense approach?
Thank you!
When developing a strategy I mostly look at ARR/MaxDD and require MaxDD < 10% with at least 1000 trades over each 5y internal. After looking at thousands of strategies I have decent idea on what capital requirements would be for different ones and I simple drop the ones that would use too much capital (let's say anything >35%). As they would be very difficult to combine with others to reduce overall expected DD and increase returns.
Mostly I'll stick to those that have ~15-20% avg exposure. At the peak they might use as much as 90% but are much less on average. Individual strategies, typically, don't offer ARR/MaxDD better than 1:1 or 2:1. And even if they do 2:1 I would still expect them to have 1:1 live. To improve that I combine multiple ones with as little correlation as possible. If correlation is assessed well - that ratio for combined equity can go as high as 8-4:1. On a margin account you can combine 3-5 with characteristics mentioned above, which is typically what I have.
When trading stocks, low correlation often means having peak capital usage = most positions during different market phases. Let's say one would have most positions when market is making new highs and another at new lows, with hold period short enough to avoid them colliding. This way, by design you shouldn't have max capital use by all strategies at the same time. It gets harder as you add most strategies but is still possible.
In my particular case, since average capital use per strategy is already low, I give each strategy 100% of account but their position sizes are set to stay within 10% DD. Typically comes down to <7% per position.
Random notes looking at my trading YTD:
- Most PL this year came from shorts, ~60%.
- Positions sizes are ~30% smaller on average than in first half of 2020.
- There were no major systems changes.
- 2 new "slower" systems added for registered accounts.
- 1 previously disabled system was re-worked and re-launched. It had a completely insane backtest ARR/MaxDD of ~20%/4%, consistent over each of 6 last 5 years periods, but I always feared that with my size I will hit its' liquidity constrains, which I think I did. New version has way more conservative liquidity assumptions, still having reasonably low correlation with other long system and good ARR while MaxDD is more realistic 13% during previous bubble. After running it for a week it "feels right", previous one kept me wondering and distracting from other stuff. But for small accounts ~50k USD if would have been a hit.
- This year I saw ~worst and ~best case examples of launching new system. One immediately went in 1.5-2x max historical DD (in Feb) and another had its' biggest +day on record the day I launched it in June. This re-enforces a good perspective - (a) that we never know what's gonna happen short term with a system, but with sufficient confidence in a testing method, our job is just to keep running it unless it gets completely out of touch with historical model. (b) worst DD is always ahead
For new readers, here is a quick reminder of why live results typically don't match backtests. If you are trying to come up with a formal systematic trading method it is critical to address all of them:
- Backtest doesn't have enough trades => not statistically significant.
- Backtest is run on bad data (quality/survivorship bias)
- Backtest is incorrect. Coding mistakes
- Execution/costs assumptions are incorrect
- Backtest is done over non-representative market sample. Typically too short or/and too market specific
Agree, retail has no chance to have an execution edge. Best we can do is hope to get violently abused and that itself is a tough ask. Custom routes and flags help, depending on factors but trying to hide size as retail in illiquid markets is excruciating.
Good point.
It will degrade eventually. Perhaps not something to be concerned much when launching a strategy, as this will be an unlikely reason why a strategy won't work right out the gate, assuming testing was done on 10y+ data sample.
Before taking off for a vacation I looked at a recently launched live strategy specifically designed for a cash account and to my surprise it took off. I think it's a first time I've seen a new strategy making so much so fast.
More often than not those periods will mean revert, but as discussed in a past, there is no way of knowing so show must go on without interference.
Experience taught me to celebrate those moments, to balance out inevitable emotional lows of deep/fast drawdowns.
What a treat right before taking some rest from a very busy summer.
Celebrate you wins when you have them.
More often than not those periods will mean revert, but as discussed in a past, there is no way of knowing so show must go on without interference.
Experience taught me to celebrate those moments, to balance out inevitable emotional lows of deep/fast drawdowns.
What a treat right before taking some rest from a very busy summer.
Celebrate you wins when you have them.
- Status