Auto or manual - page 8

 
Georgiy Merts:

Whoa, whoa, whoa, whoa.

Point 3 is NOT the point. I have perfectly clear criteria for decoupling. I have difficulties with the reverse task - choosing the most stable TC. Unfortunately, I don't have this task solved, and it is performed intuitively.

Your breakdown criterion is a series of losses of some length. This information allows you to judge when this series will stop (or the next one will start) and the system will start earning again? No, it does not. So, no conclusions can be drawn from it; it is just a belated statement, like trading in the direction of the average trend. Yes, we had a trend, so what next, it will continue or will it reverse? We do not know. On what, then, are the conclusions and decisions based? On nothing, it's not an analysis, but an imitation of an analysis and self-deception.

Georgiy Merts:

And as for "the fact is inferred" - everything is simple here. The amount of losses is unprofitable only if the switch is made randomly. However, switching the TS does not have to be random. Now, even with intuitive switching, I nevertheless use quite objective criteria.

If it were "impossible to get a positive amount by adding negative systems" - I would have drained the account long ago. And I haven't, even though I've been working steadily for several years. If my principle is unprofitable - when do you think I will drain the $400 I have in my account now?

You have a random switch, it's not a pattern, is it? No. So it's random, and it's only a matter of time before I lose. Which one exactly - well, how can you know? But with your inputs (random selection from a set of deliberately draining robots) it will happen one day anyway. In the picture in the last post, as you can see, there are weakly successful simulated implementations that turned out to be in the positive area, apparently this is your case when a randomly losing system turns out to be profitable. But 1) why do we need such an inefficient trade? no losses, but also no profits, while time and effort are wasted 2) how can we seriously rely on randomness where the probability of success is knowingly and significantly lower than the probability of failure? Risk is necessary when it is justified, otherwise it is simply foolishness and relying on chance.

 
Georgiy Merts:

In other words, have you made all the money in the world already?

In 10 years, if you make at least 30% a month, how much should it be?

Or is "profitability" measured in five percent a year?

30% a month is for dreamers. There are no algorithms that yield such returns. We should just forget about such figures forever! For those in the know, it is obvious that the profitability of any algorithm rests either on liquidity or on the capabilities of the theoretical model. If you take HFT and arbitrage, you can essentially provide any yield there, even 1000% per month, but the question is the amount you can wrap. Ultimately, in these algorithms, the returns are hinged on liquidity.

If you use other algorithms, you won't get the same returns. To start with, we need a normal theoretical model and build on it.

Yes, I have achieved 20-25% annual return in foreign currency on American equities and I am working on reducing resource consumption, simplifying the algorithm, improving the yield to 30-35% per annum and smoothing out the yield curve. And this is an excellent yield, it is enough for me stably with the risks tending towards zero.

30% per annum with minimal risk - this is the grail.

 
vladavd:

Your breakdown criterion is a series of losses of some length. Does this information allow you to judge when this series will stop (or the next one will start) and the system will start earning again? No, it does not. So, no conclusions can be drawn from it; it is just a belated statement, like trading in the direction of an average trend. Yes, we had a trend, so what next, it will continue or will it reverse? We do not know. On what, then, are the conclusions and decisions based? On nothing, it's not an analysis, but an imitation of analysis and self-deception.

No, it is not. A breakdown criterion is exceeding any of a set of criteria. A series of losses is only one of them. The TS is headed for overoptimization if

  1. A series of continuous SL exceeds the maximum one recorded on the history (or an unacceptable SL has been taken, for example, a rollover).
  2. The price slippage exceeds the maximum on the history.
  3. The expectation of the maximum update exceeds the expectation on the history.
  4. The expectation of another trade exceeds the maximum on the history.

These, in my opinion, are quite sufficient criteria to remove the TS from the trade.

At the same time, there is no sense to say "in case the trend continues". If a situation is detected in the real trade, which was not observed in the entire history - it means that the TS has changed its behavior, and it does not correspond with the market, does not correspond with its trading history. It should be removed from the trade and immediately re-optimized.

It's more complicated with the setting of trades, as I said above. Many criteria are used, but they are used intuitively, simply by evaluating "by eye". So far I have not been able to formulate a clear criterion.

 
Vladimir Baskakov:
You can drive with lots indefinitely, and it's no use, there's no profit.

what do locs have to do with it? There are no lots, it's a picture in the terminal. There is only an open and a closed position.

Here is an example on 28 stocks without leverage, netting, with commissions


 
vladavd:

You have an accidental switch, it's not a pattern, is it? No. So it's random, and it's only a matter of time before it flushes. How do you know exactly which one? But with your inputs (random selection from a set of deliberately draining robots) it will happen one day anyway. In the picture in the last post, as you can see, there are weakly successful simulated implementations that turned out to be in the positive area, apparently this is your case when a randomly losing system turns out to be profitable. But 1) why do we need such an inefficient trade? no losses, but no profits either, while time and effort are wasted 2) how can we seriously rely on randomness where the probability of success is knowingly and significantly lower than the probability of failure? Risk is necessary when it is justified, otherwise it is simply foolishness and relying on chance.

No. Rotation is a natural process. The moment when the systems switch over is clearly known - as soon as one of the working systems has produced an unacceptable parameter. The number of working systems is determined by maximal load on the deposit (I try not to keep the margin level below 1000%).

The case when suddenly ALL the robots fail at once, of course, is possible, but it's also possible with any trade, whether robots or manually... So there is no difference.

About what the risk is for - I don't understand... There is risk everywhere. If you don't want to take risks, put your money in the bank.

 
Mikhail Dovbakh:

T.e. There is no way to build as reliable systems out of unreliable elements?

)

The reliability of a system cannot be higher than the reliability of the worst element.

If the elements in the system are arranged in series.


But this is not the case in a parallel system. =)

 
Maxim Romanov:

Yes, I now have achieved returns of about 20-25% per annum in currency on American equities and am working on reducing resource consumption, simplifying the algorithm, increasing returns to 30-35% per annum and smoothing out the graph of returns. And this is an excellent yield, it is enough for me stably with the risks tending towards zero.

30% per annum with minimal risk - that's the grail.

It's almost the same as a deposit in a bank... Only you don't have to make it, work... And it's much more reliable... Without any drainage...

We're talking at least 100% per annum... With such interest rates, there can be no risk-free robots... I've already said it above...

 
Georgiy Merts:

It's almost the same as a deposit in a bank... Only you don't have to make a claim, work... And it's much safer... Without any drainage...

We're talking at least 100% per annum... There can't be no risk-free robots with such interest... I've already said that above...

A deposit in a bank at 6% in rubles and 0.5-07% in foreign currency...

To come down from the heavens closer to earth and reality

Bridgewater Associates , the largest fund, shows an average return of 22.3% p.a. over 20 years. It's a top fund, it has access to the best mathematicians, the best programmers and the best minds. And it has over $100 billion in assets under management. Probably all these investors are fools to bring their money at such a low interest rate. You have to try to at least reach such figures to start with.

And why can't there be risk-free trading robots? If you trade without leverage, only on your own money, stocks from sp500, the risks are much lower than just putting money into Sber, we have a lower credit rating than these companies and so does Sber. At most what can happen is that one of the companies goes bankrupt. And if there are 500 of them in the portfolio and adequate diversification logic?

Yes, you can make more interest, you can arbitrage futures against the spot, at least trade for 10000 roubles, you can easily make 100% per month from 10000 roubles, you can make 30 exactly.

 
Georgiy Merts:

No. Rotation is carried out in a pattern. The moment of system switching is well known - as soon as one of the working systems has given an unacceptable parameter. The number of working systems is determined by the maximum load on my deposit (I try not to decrease the margin level below 1000%).

Do you not see the difference between a pattern and just some conditionality? Regularity, by definition, allows you to judge the future with a certain probability other than 0.5, if you do not have such judgments, then there is no regularity, and the criteria for making decisions - just a fiction. Therefore your decisions are random, although formally they are conditioned by something. This is logic of the level of "if you see a black cat, you are in trouble", it is irrational and absurd in the absence of correlation between the phenomena.

 
Maxim Romanov:

not ANYTHING. My last robot with the same settings has consistently traded on 28 currency pairs over 10 years, 56 US stocks over 5 years, 28 Russian stocks over 8 years and 17 cryptocurrencies over the last 3 years. Tested a total of 129 trading instruments, the equivalent of 835 years of single instrument testing.

Not bragging about the results, just that not every robot has periods of earnings.

I think the point is that any robot has its final earning and losing periods))). You certainly do not have any and the result of long and painstaking work. The problem is that the boundaries of states are blurred and the task of automating the identification of these boundaries in real life has not been solved yet... Although what I'm talking about and to whom, you already know that))))