A balance between QUALITY and QUANTITY - page 3

 
Diamant: ... But don't you think that out of these 3 results ALWAYS none of them can be regarded as at least somewhat satisfactory ...?

Of course you are. I'm not going to make a choice. I'm just improving the EA step by step.

By the way, the EA has disabled the possibility to make short operations, for now only buying, also with selling - it gets better, and in general 90-95% of its "power" is disabled, because the capabilities of the tester do not allow.

 

Here is a piece of my personal correspondence with the developers.

About money management. Here is a picture I drew.



The black line is how the price moves.

  1. Green. The ideal. any capital + ideal management (reinvestment taking into account the spread + commissions for the bribe) if the TS consists only of such deals.
  2. Red. Missed the exit point. But good too. Almost changes the money management a bit too.
  3. But the blue one is a problematic TS. It has a drawdown after entering the trade

    For it we need to think out something on condition that the LOI>1.5 on sufficient statistics. To calculate we need statistics of each transaction in pips relative to the red big points and that's it. Just calculate guaranteed result taking into account confidence interval. By the way I once asked to make a report in the tester in pips. These dollars. % only get in the way.

    When I understood that, I threw out all money management books, they are crap. The only decent one is the Pontryagin's maximum principle where everything is beautifully and mathematically shown. How to manage. What to manage and what you need to know to do it http://abitur.bsuir.by/eumk/smssu/lecture/theme_4.html

    But it is like bais everyone knows it (or has heard of it), but it is almost impossible to apply it in practice because of large and stringent requirements for a priori data.

Z.I. I'm very interested in Alexey's (mathematician) opinion. He wrote about "sandwich", that's how I see my "sandwich" (first page of this thread). If he could refute this criterion, I would be glad, therefore there is a better one

Z.U. By the way this is what you are looking for fixing a lot in the tester, if you also remove TP and SL from the trading system, it will be my proposed criterion for the best TS.

 
Prival:

Let me try to put it in other words. If you find such a TS (ideal according to my (described above) criterion), then it is the Grail, in its purest form. You can enter the trade up to your tomatoes with the whole deposit.

You have to strive for the ideal, for risk-free trading.


Then you need a market model to achieve this goal and the axiom of random walk is not appropriate at all, rather it is appropriate, but for institutional traders who perform other tasks and are almost constantly in the market, they do not need more accuracy. I have even seen a branch where they experimentally found out that price flow differs from white noise, so I think it is not a dead-end idea to develop in this direction.
 
gip: ....... I visually check the smoothness of the equity curve as it gives us a general idea of how the groups of trades are distributed; the distribution must be even, otherwise it is a fitting again.


How do you express the smoothness of a curve mathematically? I'm looking at it too.

How about this quality criterion:

Quality criterion = Smoothness of the curve [???]* Number of transactions during the test [pcs] * Duration of test [months]

or like this:

Quality criterion = Smoothness of curve [???]* Duration of all orders [months]

 

Generally speaking it depends on the strategy. The trades are clustered in their working areas, say for a flat strategy in corrections. So, to estimate these areas we need to diagnose them and then estimate the uniformity of distribution within these areas and outliers. That's why I formulated it in such a general way.

The depth of history should be the maximum at which the exploited dependence is preserved. At least two years I think.

 

Prival, your criterion for quality is clear. I'll try to say something.

Such a system seems to me to be one of the most optimal ones myself. Unfortunately, so far I've only heard of something similar (no drawdowns on entry) from one person - IgorM. His system, I assume, was multicurrency.

But Igor, as I understand it, had another problem that he still hasn't solved - the problem of exiting a trade. It is a problem on the same scale as the problem of entering a trade.

I have been working on my own system for several weeks, which, by the way, is also multicurrency. But there is some strange mathematics (more like physics) I have not understood yet :).
 
Mathemat:

Prival, your criterion for quality is clear. I'll try to say something.

Such a system seems to me to be one of the most optimal ones myself. Unfortunately, so far I've heard of something similar (no drawdowns on entry) from only one person - IgorM. His system, I assume, was multicurrency.

But Igor, as I understand it, had another problem that he still hasn't solved - the problem of exiting a trade. This is a problem on the same scale as the problem of entering a trade.

I myself have been turning my system around in my head for a few weeks now, which, by the way, is also multi-currency by design.

It can't be. - Because it can't be.

(Chasing such a TS will result in shortening trades to the spread)

 
Mathemat:


I'll let you in on a secret - the system is multi-tick)) - it is simply possible to correctly collect and filter ticks for a successful entry into a trend. Entrances by the same indicators - wipers, stochastics - in fact, using ticks you can form a rank-bar

And outputs turn out to be as important as inputs

 
Richie:


And how can the smoothness of the curve be expressed mathematically?

see the LR Correlation (Linear Regression Correlation Coefficient) indicator for the balance curve in the Championship section of the Reports tab:

 
Prival, your criteria contradict the theory and practice of an efficient market. Any open market is efficient enough to prevent your approach even theoretically, let alone practically. In other words, you might as well have spent your life creating an engine with efficiency >=100%, the engine would never have been created and your life would have been wasted.