Discussion of article "Extract profit down to the last pip" - page 22

[Renat Akhtyamov]

fxsaber:

I waited. The trade is halted.

What do you mean, who stopped it?

[Enrique Dangeroux]

every tick in MT4.

Do you believe this negatively impacted performance? Considering that you were working with some synchronisation mechanism.

Do you believe this negatively impacted performance? considering you ran with some synchronizing mechanism.

[fxsaber]

Enrique Dangeroux:

Do you think this had a negative impact on performance? Given that you were working with some sort of synchronisation mechanism.

No. It doesn't seriously affect synchronisation. Objectively, the EA started to perform at a loss. This is what the Tester shows.

Unfortunately, I have not been able to write a criterion that would show a significant difference between the unprofitable October and the profitable months before that.

The weighted average spread, potential profit and other characteristics did not show significant differences from the previous indicators.

I really hope to create a criterion that would show the difference between October and other months. Of course, the Expert Advisor itself cannot be such a criterion.

[Enrique Dangeroux]

Around 2018-10. It was also an unfavourable period for the algorithm. I wouldn't be surprised if this is a similar period and if it works for a while, the algorithm will work again.

I hope you can find an explanation.

Around 2018-10 there also was unfavourable period for the algorithm. I would not be surprised if this is similar period and given some time, the algorithm will perform profitable again.

I hope you will find explanation.

[fxsaber]

How markup affects profit.

In theory, in case of negative markup (prices improve) at the same TS settings, its expectation should increase by double markup if trading signals are repeated.

For example, there was a profit of 20000 pips at 1000 trades. We made markup -1 pip (Bid -= -1pips, Ask += -1pips). Then MarkupProfit = 20000 - 1000 * (-1 ) * 2 = 22000 pip.

But markup affects the prices, on which the trading logic depends, so the signals will not be repeated. Moreover, when optimised, the TS should adjust even better. Because in the example above, a pass must be found that has a profit of more than 22000 pips with a larger number of trades. I.e. the expectation will decrease (because it is more profitable to catch smaller movements), but the profit will increase.

This is all in theory. But how to do it in practice? MT5 makes it very easy to conduct such research.

So, seven symbols are created with the following markups: -3, -2, -1, 0, +1, +2, +3 pips. The source material is a real symbol. The zero markup is the same without modifications.

Optimisations were performed on each - seven pieces. The best settings were taken from each optimisation and a run was made with them on all symbols. Thus, the following table was filled in.

Result

What markup corresponds to best settings	Markup -3 pips	Markup -2 pips	Markup -1 pip	Markup 0 pips	Markup +1 pip	Markup +2 pips	Markup +3 pips
Markup -3 pips	Profit: 18891 Trades: 1348 MO: 14.01 PF: 2.38 Profit0: 10797	Profit: 16165 Deals: 1198 MO: 13.49 PF: 2.21 Profit0: 11369	Profit: 14904 Deals: 1067 MO: 13.97 PF: 2.15 Profit0: 12771	Profit: 12686 Deals: 897 MO: 14.14 PF: 2.03 Profit0: 12686	Profit: 10255 Deals: 757 MO: 13.55 PF: 1.87 Profit0: 11771	Profit: 8728 Deals: 647 MO: 13.49 PF: 1.77 Profit0: 11316	Profit: 5759 Deals: 547 MO: 10.53 PF: 1.50 Profit0: 9041
Markup -2 pips	Profit: 16430 Deals: 1089 MO: 15.09 PF: 2.27 Profit0: 9899	Profit: 16832 Deals: 1045 MO: 16.11 PF: 2.37 Profit0: 12654	Profit: 14893 Deals: 938 MO: 15.88 PF: 2.25 Profit0: 13019	Profit: 11478 Deals: 812 MO: 14.14 PF: 1.97 Profit0: 11478	Profit: 9110 Deals: 705 MO: 12.92 PF: 1.79 Profit0: 10518	Profit: 7386 Deals: 618 MO: 11.95 PF: 1.66 Profit0: 9857	Profit: 6333 Deals: 547 MO: 11.58 PF: 1.59 Profit0: 9616
Markup -1 pip	Profit: 15566 Deals: 1863 MO: 08.36 PF: 1.84 Profit0: 4396	Profit: 14696 Deals: 1588 MO: 09.25 PF: 1.87 Profit0: 8337	Profit: 14806 Deals: 1313 MO: 11.28 PF: 2.01 Profit0: 12184	Profit: 10603 Deals: 968 MO: 10.95 PF: 1.80 Profit0: 10603	Profit: 9422 Deals: 752 MO: 12.53 PF: 1.79 Profit0: 10926	Profit: 6506 Deals: 591 MO: 11.01 PF: 1.55 Profit0: 8870	Profit: 7129 Deals: 494 MO: 14.43 PF: 1.7 Profit0: 10092
Markup 0 pips	Profit: 16033 Deals: 1457 MO: 11.00 PF: 2.20 Profit0: 7285	Profit: 15279 Deals: 1350 MO: 11.32 PF: 2.16 Profit0: 9882	Profit: 13638 Deals: 1217 MO: 11.21 PF: 2.05 Profit0: 11208	Profit: 13137 Deals: 1068 MO: 12.30 PF: 2.12 Profit0: 13137	Profit: 12244 Deals: 952 MO: 12.86 PF: 2.10 Profit0: 14146	Profit: 10415 Deals: 822 MO: 12.67 PF: 1.95 Profit0: 13702	Profit: 8815 Deals: 712 MO: 12.38 PF: 1.83 Profit0: 13086
Markup +1 pip	Profit: 14246 Deals: 1219 MO: 11.69 PF: 2.04 Profit0: 6936	Profit: 13248 Deals: 1026 MO: 12.91 PF: 2.08 Profit0: 9141	Profit: 11464 Deals: 818 MO: 14.01 PF: 1.97 Profit0: 9824	Profit: 12197 Deals: 679 MO: 17.96 PF: 2.21 Profit0: 12197	Profit: 10834 Deals: 655 MO: 16.54 PF: 2.07 Profit0: 12143	Profit: 9331 Deals: 627 MO: 14.88 PF: 1.91 Profit0: 11837	Profit: 7971 Deals: 540 MO: 14.76 PF: 1.80 Profit0: 10130
Markup +2 pips	Profit: 15617 Deals: 1152 MO: 13.56 PF: 2.32 Profit0: 8709	Profit: 12717 Deals: 942 MO: 13.50 PF: 2.08 Profit0: 8949	Profit: 11016 Deals: 732 MO: 15.05 PF: 2.04 Profit0: 9552	Profit: 11617 Deals: 575 MO: 20.20 PF: 2.31 Profit0: 11617	Profit: 10923 Deals: 565 MO: 19.33 PF: 2.22 Profit0: 12051	Profit: 9876 Deals: 551 MO: 17.92 PF: 2.08 Profit0: 12077	Profit: 7484 Deals: 444 MO: 16.86 PF: 1.86 Profit0: 10149
Markup +3 pips	Profit: 16626 Trades: 1491 MO: 11.15 PF: 2.17 Profit0: 7678	Profit: 13294 Deals: 1176 MO: 11.30 PF: 2.03 Profit0: 8584	Profit: 11486 Deals: 913 MO: 12.58 PF: 1.96 Profit0: 9659	Profit: 11271 Deals: 713 MO: 15.81 PF: 2.10 Profit0: 11271	Profit: 9763 Deals: 682 MO: 14.32 PF: 1.94 Profit0: 11130	Profit: 8234 Deals: 651 MO: 12.65 PF: 1.78 Profit0: 10839	Profit: 8662 Deals: 517 MO: 16.75 PF: 1.91 Profit0: 11761

For example, cell (column; row) = (-2; +1) contains this data: an optimisation was performed for the +1 character and the best result settings were applied to the -2 character. I haven't seen a study like this anywhere, so it was very interesting. So, seven optimisations (for each character) and for all optimisations seven single passes.

Applied a multitester for full automation. The manual fiddling is filling out this table.

Profit0 is the theoretical profit on a real symbol, if the trading signals matched. The highlighted piece in the table is the best Profit0.

Conclusions.

Specially brought a lot of numbers so that you can see some "laws". It is clear that the study is not exact, because genetics was used during Optimisation. But roughly something can still be deduced.

• As a rule, the more you worsen the prices for optimisation, the corresponding better pass on the real symbol gives higher PF and average deal. This is due to the fact that the number of trades drops a lot more than the profit.
• It is possible to improve profits on a real symbol if you optimise on a degraded symbol by substituting the trading signals of the better pass into the real symbol. But this is probably an exception to the rule (need to check).
• Basically, the theory is confirmed that optimisation on a native symbol almost always gives a better result than trading signals of the best passages of unmarked symbols.
• To study the TS, it makes sense to run it on off-markup symbols.
• The highlighted diagonal in the table perfectly demonstrates how much profit changes when the markup changes by just 1 pip. It is a good counter-argument for those who think that price difference by 1-2 pips does not seriously affect anything.

[Denis Kirichenko]

Colleague, very interesting research!

But when comparing, imho, some values that form a set of values, it is necessary to use statistical methods.

As I understand, in this case the task is to determine whether the markup affects the performance of the system.

Then:

• H0 (null hypothesis): the results with and without the markup are equal. I.e. the addition of the markup has no effect on the system performance.
• H1 (alternative): not equal. I.e. the addition of markup is of fundamental importance in determining the profitability.

I would also add that this:

Но маркап влияет на цены, от которых зависит торговая логика, поэтому сигналы повторяться не будут...

will be levelled by the multiplicity of the number of elements of the population. I.e. it will be possible to compare two sets of results.

This is the opinion.

[fxsaber]

Denis Kirichenko:

When comparing, imho, some values forming a set of values, one should use statistical methods.

In this case, I don't quite understand how to statistically compare optimisation clouds obtained through genetics. Besides, the TC from the article is used here, and this is a very special case of TC.

As I understand, the task in this case is to determine whether markup affects the performance of the system.

The original task was different. It was to understand whether it is normal or not, when markup affects the entry points in the strongest way? Such a dependent trading logic is not very favourable to me. Unfortunately, the TS from the article is exactly with such logic.

ZЫ I added a point to the conclusion

Forum on trading, automated trading systems and testing trading strategies.

Discussion of the article "Scratching the profit to the last pip"

fxsaber, 2019.11.09 15:54

• The highlighted diagonal in the table perfectly demonstrates how much profit changes when the markup changes by just 1 pip. A good counter-argument for those who think that a 1-2 pip price difference doesn't seriously affect anything.

[Denis Kirichenko]

fxsaber:

In this case, I don't quite understand how to statistically compare optimisation clouds obtained through genetics. Besides, the TC from the paper is used here, and this is a very special case of TC...

You can compare populations. Let's say there is sample 1 and sample 2. A statistical test can tell if they're the same or not.

When genetics calculates some combination of parameters at markup=0, and skips (nullifies) it at, let's say, markup=1, it is natural to compare such populations with a reservation. It is better when we compare the same combinations of parameter values at different markups.

The initial task was different. We had to understand whether it is normal or not, when markup has a strong influence on entry points. Such a dependent trading logic is not very favourable to me. Unfortunately, the TS from the article is exactly with such logic.

Well, that's what I wrote about hypotheses. I'll paraphrase:

• H0 (null hypothesis): markup does not affect the result of the trading system.
• H1 (alternative): markup affects the result of the trading system.

If the last thesis is confirmed, then the system is sensitive to additions to the initial price. And it is rather bad.

[fxsaber]

Denis Kirichenko:

It is better when we compare the same combinations of parameter values at different markups.

The table does just that. Each row has the same set of inputs.

Well, that's what I wrote about hypotheses. I'll rephrase it:

• H0 (null hypothesis): markup does not affect the outcome of the trading system.
• H1 (alternative): markup affects the result of the trading system.

If the last thesis is confirmed, then the system is sensitive to additions to the initial price. And it is rather bad.

It can be seen that in each line the profit falls from left to right, which is fully consistent with the theory. Moreover, the number of trades also falls in the same direction. Which just speaks in favour of the fact that the inputs are not taken the same, but more OPTIMAL. Well it's strange when the inputs don't change during markup. Here is the bottom line

Forum on trading, automated trading systems and testing trading strategies.

Discussion of the article "Scratching out the profit to the last pip"

fxsaber, 2019.11.09 15:54

Which markup corresponds to best settings	Markup -3 pips	Markup -2 pips	Markup -1 pips	Markup 0 pips	Markup +1 pip	Markup +2 pips	Markup +3 pips
Markup +3 pips	Profit: 16626 Trades: 1491 MO: 11.15 PF: 2.17 Profit0: 7678	Profit: 13294 Deals: 1176 MO: 11.30 PF: 2.03 Profit0: 8584	Profit: 11486 Deals: 913 MO: 12.58 PF: 1.96 Profit0: 9659	Profit: 11271 Deals: 713 MO: 15.81 PF: 2.10 Profit0: 11271	Profit: 9763 Deals: 682 MO: 14.32 PF: 1.94 Profit0: 11130	Profit: 8234 Deals: 651 MO: 12.65 PF: 1.78 Profit0: 10839	Profit: 8662 Deals: 517 MO: 16.75 PF: 1.91 Profit0: 11761

Looking at the extremes. Markup differs by 6 pips. At the same time the number of trades differs three times.

Let's see what the profit would be if we trade on the right cell: (16.75 + 6 * 2) * 517 = 14863, which is clearly less than 16626 in the left cell. At the same time, look at the expectation in the left cell, it is less than double markup! I.e. everything is very logical. The TS caught a lot of small fluctuations, while the input parameters corresponded to the best pass on a very marked symbol. Such a property of the TS should not upset. But there are still some doubts.

In principle, this is probably how any TS behaves, but we are not sure about it.

[fxsaber]

fxsaber:

• It is possible to improve the profit on the real symbol, if you optimise on the degraded symbol, substituting trading signals of the best pass into the real symbol. But this is most likely an exception to the rule (you need to check).

I made a check through genetics using this criterion

sinput int inMinTrades = 0; // Minimum number of trades (positions).

double OnTester()
{
  return((TesterStatistics(STAT_TRADES) > inMinTrades) ? (TesterStatistics(STAT_EXPECTED_PAYOFF) + GetMarkup(_Symbol) * 2) * TesterStatistics(STAT_TRADES) : 0);
}

I.e. I searched for the maximum profit for a real symbol, if the trade entries coincide. Here are the best passes for each symbol

Which markup matches best settings	Markup -3 pips	Markup -2 pips	Markup -1 pip	Markup 0 pips	Markup +1 pip	Markup +2 pips	Markup +3 pips
	Profit: 14300 Trades: 512 MO: 2793 PF: 2.70 Profit0: 11228	Profit: 17719 Deals: 1169 MO: 15.18 PF: 2.21 Profit0: 13043	Profit: 13959 Deals: 601 MO: 23.23 PF: 2.66 Profit0: 12757	Profit: 13137 Deals: 1068 MO: 12.30 PF: 2.12 Profit0: 13137	Profit: 11449 Deals: 1052 MO: 10.88 PF: 1.82 Profit0: 13553	Profit: 9891 Deals: 839 MO: 11.79 PF: 1.83 Profit0: 13247	Profit: 7746 Deals: 876 MO: 8.84 PF: 1.59 Profit0: 13002

It's clear that it's genetics. And, for example, the maximum that was in the original table was not found, which raises some questions for the GA...

However, we can conclude that it probably does not make sense to consider trading entries from deteriorated symbols rather than to increase the profit, but to increase the mat expectation.