Is this even possible?
The liquidity is fine
What happends when you up the number of holdings to 30 - 50 - 100?
1 Like
Looks good. How large is the universe you are using?
1286 Now
That's why I didn't try a 100-stock variant.
Edit:
I tried lowering the excessive share turnover and somehow the returns became higher.
Assuming 5% of trading volume each day, you're expecting these stocks trade $4m a day ($200,000 positions). Try it on small caps and see how the results look?
The minimum 20% iquidity is $5m+ so you can invest $10m+
I included a portion of mid-cap stocks because there aren't enough small-cap stocks with enough liquidity.
Edit:
With the addition of a 10% market cap factor, it appears that the results gained significant improvement.
The thing goes even wilder when I try to improve the sell rules a little
The 3M average liquidity of most of the latest positions was greater than $1 million and the minimum 3M average liquidity was over $900k.
Edit:
Was the training period you screenshot above for the AI Factor training or for the predictor? If the former, what was the training period for the predictor?
With 1200% annual turnover, what's the Avg Return winner/losers?
Just as a point of comparison, here's a simulation starting on the same date, with weekly rebalancing, run with a ranking system I created in late 2018 and haven't touched since. No buy rules, 1 sell rule (rankpos > 40), easy to trade universe, variable slippage.
Another point of comparison: the actual unleveraged returns of my US-and-Canada long positions over approximately the same period (this starts March 20, 2019):
So, yes, trust your eyes! This kind of outperformance is indeed achievable out-of-sample.
2 Likes
Minimum 3M liquidity of the lastest holdings: > 750k
The only difference between this strategy and my previous 90%+ p.a. strategy is that I defined a less illiquid universe and retrain my AI factor
Edit:
Edit2:
Low turnover and the same AI factor trained in the US microcap universe but used in the Non-US smallcap universe
Real CPI adjusted MktCap < 8B EUR
What is the market capitalization distribution of the latest positions in your strategy
The bars correspond to up to the number in market cap, so the leftmost bar is 0 to 50 million and the rightmost bar is 10B to 20B.
@ZGWZ - One of the images above shows the rule mktcap < 8000 / (##CPI/52000) / Pow(1.025, (2024-Year)). ##CPI returns the static internal id for CPI and not the CPI value. To get the CPI value use Close(0,PITSeries(##CPI)).
Thanks
Edit1: Retraining using modified universe and low-frequency targets/features
Edit2: After increasing the liquidity requirement:
Edit3: Low turnover and high liquidity in the US microcap&smallcap universe with only liquidity screen
So you can also profit a lot with much liquidity and low trading commitments in the US market