All three across the whole system. Callbacks and coroutines just for the API, although coroutines only used for walk-forward optimization which I'm not using at the moment.
As I was expecting, here came a couple of big losers. The first one (-864$) was my mistake. I had to stop the system for a little maintenance and during that time an entry signal would have been triggered. So when the bot was online again, about half an hour later, the price was still going in the right direction so I decided to open the trade anyway. Well deserved, for remembering not to chase price. Had I opened the trade at the original signal it would have been a winner (+610$). This is the main reason why I strive to remove myself as much as possible from the actual trading. Still not enough, apparently. After that I decreased the position size down to two contracts. The second loser (-819$) was in line with the backtest, as per algo. Decreased position size again. I will be increasing it back again when things get better.
Just jumping in to say that this is a great thread. I recently went live with my first automated/algo strategy (after much paper trading, backtesting, and even more testing the backtesting), and it's been fun -- and so far, profitable -- process. You learn a different facet of how markets work when you're forced to think about them programmatically instead of from a human decision-maker's point of view. My two takeaways so far have been that t's better to over-fit than to under-fit (within reason), and "automated" doesn't mean "hands-off."
Yes, I think so. I wrote something above on fitting and overfitting. I stick to getting as many trades as possible from the backtest, at least 1.000 for the period of interest. However even that is not a guarantee of any sort. No more trades for the moment, bot currently on the sideline.
Why? I would have been interest to hear how it performs in a correction like we had in the last 2 weeks.
The system trades mostly long in presence of an overall uptrend and mostly short when in a general downtrend. However, since my tests have shown more volatility and lower reliability on the short side, I designed the short algos in such a way that they are more "cautious" than those on the long side. So there can be relatively extended periods, like the current one, during which the system is flat.
So this means that your system is still running, but because of the current state of the equity market it has no open positions? I must have misread your previous message as I got the impression that you had switched it off.