Is it better not to trade at 11 p.m.? - page 7

New comment
 

zaskok3:

I haven't counted the turnover, but it's obviously huge.

I had a turnover of $56 mio in four days of not fully round-the-clock trading. He traded around the clock, more often than not, 10-12 times as much. So in a week his turnover is roughly $1 billion.
 
zaskok3:
I had a turnover of $56 mio in four days of not fully round-the-clock trading. He traded around the clock, more often than not, 10-12 times as much. So in a week his turnover is roughly $1 billion.
Fifty-six million?
 
Alexey Busygin:
56 million?
What a, Ostap Benderbin oglu bens.
 
This is a turnover of 560 lots.
 
The goal is not to repeat the TS. But, in general, the trade turns out to be similar. Apparently, even if the TS is based on different logics but shows profit, the trading signals must correlate quite strongly. The statement is not obvious at all. But perhaps since both TS use the same market pattern, it is characterized by a certain set of entries.

A hypothesis can be deduced from it:
If two TS with different logic are strongly correlated by inputs, then they use the existing market pattern.

Overnight there should have been a flip of SELL-positions in the same place where the guy was flipping at the same time. My lots were shallower and the number of orders was much higher.

The guy flipped, I almost didn't. In the morning instead of plus, as the tester showed, I got corresponding minus.

Since the battle robot trades on the principle of synchronization of the battle trading environment with the virtual tester one, it became apparent that the synchronizer does not account for such an event when a rollover has not occurred and there is no second chance (a rollover at a local extremum).

I found the solution. It turns out that it is on the real market (not in the tester) where we should separately consider the logic of setting opposite Limits and Take open positions. In short, the reversal TS for the tester should not use the reversal logic for the real trade.

I'm sure the guy doesn't take that kind of approach. He doesn't bother - he just trades profitably.


 
So grid + averaging. Dangerous naka.
 
Heroix:
That's a turnover of 560 lots.
One billion more and one less
 
Looks like the lad has doubled his lots... Before that EURCHF was heavy in terms of channel trading (my TS only held at zero). But Paranek made a plus there as well. All in all, so far the benchmark channeler.
 
zaskok3:
Looks like the lad has doubled his lots... Before that EURCHF was heavy in terms of channel trading (my TS only held at zero). But Paranek made a plus there as well. So, so far, it is a masterpiece of channel trading.
Well, what did you expect, that it would be negative?
 
Observations of the guy's trading have led me to doubts as to whether my own approach to composing a TS combat portfolio was correct.

I will first write how I did it myself. So I have a TS. In this case it is a channeler. I bruteforced it in the tester. Threw out, subjectively, not robust variants. Then I sorted the remaining options by profit (FF, PV) and decreased the profit (FF, PV) and selected those that were as much different as possible based on values of input parameters.

Such a porftel of TC seemed to me the most differs for combat purposes. I did not bother choosing a weighting coefficient in a portfolio for each TS and simply divided the risk equally between them. Nor did I build their combined equity.

It seemed to me that this approach was somewhat logical. However, when comparing my TC portfolio and the guy's TC portfolio, I saw differences. Very often it happened that seemingly different TSs in the portfolio turned over in almost the same places. But in the kid's case it was happening differently.

It felt as if his portfolio consisted of short-term and long-term TS. When a frequent channeler went into minus, a rarer one pulled out with its own profit. And vice versa. I.e. we can see a completely different approach to creating a portfolio of TPs - the "nettler". But not the classical understanding of the nettler, that's why it is in quotes.

Therefore considered (only theoretically so far) another approach. If there is a task to make a portfolio of TSs with different planning horizons, we should take TSs with different MIs after bruteforcing in the optimizer and different sorting.

Suppose we've decided upon the range of acceptable values of IL. Then we divide this interval into several equal sub-intervals and for each one we take one variant of TC falling to this sub-interval.

There is one more idea related to building the smoothest portfolio equity curve on the history. I.e. we find such a porftel so that equity would be as nice as possible. In this case, the portlet should automatically include short- and long-term TPs which will insure each other. But it is not a fact.

In general, such equity smoothing tools may be applied directly (to bruteforce) or to already selected parameters (by CB, as I wrote above) to find weights for each TS in the portfolio.

If I'm not mistaken, Markowitz's portfolio algorithm is relevant to solving such a problem. And so I wonder how such a problem is set (equity smoothing) and solved in matrix packages. I think it's some elementary functions that were implemented a long time ago. Someone must have used them and it would be good if they told me.

So far I've come up with (not tried out) such an algorithm. Subtract the profit component (either LR or another variant) from BP of each TS's euthyte. Then obtained BPs are added together so that the resulting line is as horizontal as possible - minimization of dispersion. Fortunately, portfolios with minimum variance have been built.

I may have painted for myself a somewhat idealised image of the guy trading and he is trading in a completely different way. But I have a thought, which is not a bad thing. I admit that some of the trades in the guy's portfolio are loss-making or are near zero. But they allow equalizing equity. Although I have not yet understood the reason why it is possible.

UPD: I have moved it to a separate thread.

12345678910
New comment