(Successful) automated trading is not easy, but possible. And yes, it can be a lot of fun to watch your bot operating. As for the discretion part, in my case it is limited to pausing/resuming the strategies during periods of very high volatility (occurred two or three times in a year). The strategy parameters are static, no reoptimization. My thread/journal: I wrote a bot and gave it 25k to trade.
Thank you for your replies. @ Vtechno Thanks for reading the full post. Yes, as I did say at the beginning, this is my first experiment in a Forum and as an added experiment I thought it would be interesting to see if there are major differences between forums. This one is winning in terms of replies, whatever that means. I should post it in many others to have more meaningful statistics but that would be too much work even for me and even in a lock down @Butterfly Thanks for taking the time to post your insightful comment. I will let you know if I open a Twitter account. In the meantime, let me try to help you: my experience: purely discretionary perhaps sucks, purely automated perhaps sucks, semi-automated perhaps good. Or, if you prefer: PDPS,PAPS,SAPG @ terminus Very interesting thread. I am afraid that your technological skills are above my league. Very impressed, though. @ValeryN Thanks. My wife appreciates. I actually thing that a supportive partner is one main requirement in an independent trader’s endeavour. Very interesting thread. Really impressed again. In general: I have read all of Robert Carver’s books and I am a big fan. Knowing his views, I guess Terminus’ SR of 2,67 would make RC frown
Well, it did not take long for this thread to become moribund. Was it something I said? I reckon this is not exactly exciting stuff. Still, I would like to try to re-invigorate it (yes, I do not really have much of a life these days!) I will start with an example on the importance of measurement and stats. Probably this is more indicated to more discretionary traders, semi mechanical traders and all those who do not keep a trading journal. But there is a bit for everyone, I hope. Here is an example from one of my systems. I will use the Time of the Day (London based) stats. I have divided the stats into two periods: Sep-Dec 2020 and Jan-Feb 2021 so far and on two different trade management strategies (series 1 and series 2.) This is just an example. Before you complain: my stats are over a much longer period. Based on the 2020 stats, there is no reason to think that the system is time of the day sensitive. There is good performance in the morning, an irrelevant dip just before the US open, and a good performance again in the afternoon. In 2021 something has changed quite dramatically. The afternoon performance is consistent both in terms of time and strategy, but the morning one is all but gone. Only the 11-12 slot is OK- ish, but its best performance is with a different strategy than in 2020. There are several observations we could make from this simple example: a) If you do not have a journal, it would take you much longer to realise what’s going on where and the extent of it. I am pretty sure that in these circumstances many discretionary traders will find that their methodology (or their magic) does not work as it used to and scratch their heads. b) Systems are blunt instruments: rules are the same for all market circumstances, but obviously some circumstances are better than others. You have to be consistent. You should not pick and choose. We said that system trading is a number game. Diversification (even across time of the day) is a usually a good solution. Imagine if, based on the 2020 stats of this example I had elected to trade this system only in the mornings, for whatever reason. I would declare that “the system is broken” and look for something else. Had I elected to trade only in the afternoons I would have a totally different perception. Still, I can’t be sure that next month it will not be the opposite. c) It would be very tempting at this point to start fiddling with the parameters. The risk of overfitting is always around the corner, though. The more general the rules, the more consistent the stats. d) If your system is based on ideas that you do not understand or accept, you would have big doubts by now and would be much more inclined to quit. To better illustrate this point: I discussed half-jokingly with one reader about a rule based on the correlation between the neighbour’s dog barking 5 times and the S&P rising 1%. Such system would be based on a ridiculous theory that I can’t possibly accept. Or perhaps the neighbour moved or the dog died. So, I would now be quite ready to quit the system. e) On the other hand, if I know and accept the ideas underlying the system, I may have a better understanding of what’s going on and probably a better assessment of what (if anything) to do. This system is built on the basis of momentum and break-out. It is a well known and tested theory that resonates well with my beliefs. So, I know this system needs volatility to work well. Volatility in the European mornings have been really bad so far in 2021. I can now analyse the causes. I can decide it is just a phase and things will go back to normal in the long run. Or I can decide there is a fundamental shift in trading behaviour where European are generally sitting on the fence and wait for Americans to step in and move the markets. Perhaps I could keep trading all times, to maintain diversification, but reduce my risk based on some measure of volatility. If I were nerdy and capable enough I could try to build a systematic solution by incorporating a rule that tracks the volatility (perhaps a rolling measure of volatility, or a “volatility momentum”) and increases/decreases the trading risk accordingly. And so on. Another observation would be: when should we actually quit a system? But this is getting too long again. Ah, we always have the option of buying Tesla and Bitcoin and forget about all this, of course. Sleep well.
Hi. Long time. Not dead yet. Here is an example of the dilemma: totally automated – semi automated systems. Here is an edited excerpt from the analysis of one my systems for the month of February. When I talk of filters I mean additional rules on the basic system’s rules. Some of them could be automated, but not all, hence the “semi-automation” (read human intervention.) Against the benchmark of not filtering at all, which would have given a net of fees result of -8.7R for strategy 1 and 2.2R for strategy 2, almost all filtering would have worked. In this particular instance, taking only the “purest” trades, the ones fulfilling all the requirements given by all filters taken together, would have given the best result. In particular, look at the last line. The gross performance for the strategy 2 was basically the same (16.9R vs 17R.) So, this filtering would not have been relevant if we looked at the gross return. But we would have filtered out so many trades that the fee saving component would have contributed 9 full R! However, not all filtering rules can be coded (in particular the "PA2" filter, which is the one that is more typically discretionary, i.e. how well we read the markets with our “superior” discretional knowledge.) So, taking all filters means trading partially discretionally, which would have required us to stare at the screens throughout the trading period following every trade. Back to sleep.