Artificial Algorithmic Systems Real Trading
- Thread starter Alis
- Start date
Quick summary for August:
Starting account value was 94902 PLN, now it's 109416 and I've just paid in another 5098 PLN. That means profit for August is around 10%. Most of it came from trades in SLZ. Total number of positions now is 21. Closed positions profit: 8962, open positions profit 9273 (actually less, it doesn't include today's fall in SLZ yet).
Today's deposit brings initial cash to exactly 100k PLN. I'm not going to deposit more into this strategy as I don't want to blur PL calculations.
Starting account value was 94902 PLN, now it's 109416 and I've just paid in another 5098 PLN. That means profit for August is around 10%. Most of it came from trades in SLZ. Total number of positions now is 21. Closed positions profit: 8962, open positions profit 9273 (actually less, it doesn't include today's fall in SLZ yet).
Today's deposit brings initial cash to exactly 100k PLN. I'm not going to deposit more into this strategy as I don't want to blur PL calculations.
A bit busy day today.
First of all BAH is going bust and I force-closed all positions without waiting for an actual signal. That means systems 3892946934609429801, 420488521 and 4541748589103239761 closed all they had at 0.22 (around 50% loss).
System 2813552119 closed STS at 0.58
System 3644788620 closed SLZ at 1.45
System 9531418900514171870 bought 4615 of ASM at 0.62
System 9531418900514171870 bought 4322 of HRS at 0.692
In general it seems like I'm leaving SLZ and getting into ASM, IPO, HRS but have to wait for up-to-date signals tomorrow.
First of all BAH is going bust and I force-closed all positions without waiting for an actual signal. That means systems 3892946934609429801, 420488521 and 4541748589103239761 closed all they had at 0.22 (around 50% loss).
System 2813552119 closed STS at 0.58
System 3644788620 closed SLZ at 1.45
System 9531418900514171870 bought 4615 of ASM at 0.62
System 9531418900514171870 bought 4322 of HRS at 0.692
In general it seems like I'm leaving SLZ and getting into ASM, IPO, HRS but have to wait for up-to-date signals tomorrow.
Systems 209666634 and 2813552119 sold WIS at around 0.36-37 (order was partially filled for a few days, remaining shares sold today).
System 704827598 bought 3896 shares of ASM at 0.77. Now it's blocked trading so I suspect it will gap a lot tomorrow.
System 3892946934609429801 bought 2857 shares of IPO at 0.98
System 16952843868086350755 sold 3RG at 0.477
Account value right now is at 108k, so around 8k profit for August.
I'm having lots of signals for IPO and ASM, so probably this will be my trades tomorrow.
System 704827598 bought 3896 shares of ASM at 0.77. Now it's blocked trading so I suspect it will gap a lot tomorrow.
System 3892946934609429801 bought 2857 shares of IPO at 0.98
System 16952843868086350755 sold 3RG at 0.477
Account value right now is at 108k, so around 8k profit for August.
I'm having lots of signals for IPO and ASM, so probably this will be my trades tomorrow.
I don’t really like situations like recently with BAH where I’ve lost around 50%. In this particular case very little could have been done (market stopped trading for some time after news and then gapped) but there were (and will be) more cases like this when I will be able to exit.
A small recap before I get to testing: my systems are created and tested by a computer and I don’t really mess with them later. One of possible exits is where loss exceeds given percentage in given number of days. Stop loss exit rule can be joined with more rules. In other words, systems can get stop loss strategy “discovered” during training, and if it’s beneficial such system would win with one without a stop loss.
But still, I wanted to see what would happen if I force stop loss at an execution layer. So systems stay the same, but if loss (by EOD close price) exceeds 20% - the position will be closed.
To my surprise it’s rather disappointing. Below some data:
Original With stop loss
System count 312 67
Final account value 1188 966
Profit factor 3.54 2.48
Expectancy 1.19 0.8
Average drawdown -31 -32
First of all, only minority of systems survived trading with stop loss (67 out of 312). That means, that trading with stop loss made results (on in-sample data) so bad that most systems weren’t allowed to trade on out-of-sample. Average final account value is smaller, but given largish standard deviation there is no big significance here (two sample T test shows significance only at > 0.2). Other metrics are also worse and I wonder what could be the reason. Any ideas?
A small recap before I get to testing: my systems are created and tested by a computer and I don’t really mess with them later. One of possible exits is where loss exceeds given percentage in given number of days. Stop loss exit rule can be joined with more rules. In other words, systems can get stop loss strategy “discovered” during training, and if it’s beneficial such system would win with one without a stop loss.
But still, I wanted to see what would happen if I force stop loss at an execution layer. So systems stay the same, but if loss (by EOD close price) exceeds 20% - the position will be closed.
To my surprise it’s rather disappointing. Below some data:
Original With stop loss
System count 312 67
Final account value 1188 966
Profit factor 3.54 2.48
Expectancy 1.19 0.8
Average drawdown -31 -32
First of all, only minority of systems survived trading with stop loss (67 out of 312). That means, that trading with stop loss made results (on in-sample data) so bad that most systems weren’t allowed to trade on out-of-sample. Average final account value is smaller, but given largish standard deviation there is no big significance here (two sample T test shows significance only at > 0.2). Other metrics are also worse and I wonder what could be the reason. Any ideas?
It appears adding this check on volatility (EOD close price > 20% lower than entry price) discards too many good rules.
If you presented more detail on how the system decides to enter and exit (e.g., specific examples of rules for entry and exit), it might be possible to guess why adding stops reduces performance.
My head hurts just thinking about how you're managing that many systems. How do you actually keep track of performance for each one? You have an equity chart?
Like why do you have these long system ID? if if it's 600 systems, would'nt you just write system 1, system 2, ....system 600 bought/sold etc.
You're either a high functioning austist or a weird dude who hoards stuff.
It would be borderline impossible. I mean, systems are serialized to JSON, so they're kinda human readable, but figuring out the rules would be hours of work per system, and we're talking like hundreds of them.
In general, entry rules can be a mix of technical indicators, price action, volume, fundamental data and how all of these change in time. Entries probably differ between systems as it's required to avoid holding the same positions (to some extent obviously).
I just grepped through them and about 10% of them already use some kind of stop loss (usually in the range of -30 .. -10 %).