Discussion of article "Applying the Monte Carlo method for optimizing trading strategies" - page 3

[Aleksey Nikolayev]

Maxim Dmitrievsky:

Books? I think there's only articles

fxsaber gave arguments in his other article

I don't know about other methods... if I did, I would have already attached them :)

The question was about other methods proposed by the author of the article.

In the article about books:

1.Harris, M (2016), Limitations of Quantitative Claims About Trading Strategy Evaluation, SSRN, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2810170

2.Harris, M (2015), Fooled by Technical Analysis: The perils of charting, backtesting and data-mining, Price Action Lab. Available at http://www.priceactionlab.com/Blog/the-book/

[cemal]

Any plan to extend the article White reality? https://quant.stackexchange.com/questions/21163/whites-reality-check-for-pair-trading? utm_medium=organic&utm_source=google_rich_qa&utm_campaign=google_rich_qa

[fxsaber]

Probably, this way of evaluating the adequacy of the optimisation criterion has a right to life:

1. We optimise, for example, by R^2. We get a nice straight line of equity.
2. We take half of that in point 1 as the testing interval.
3. Optimise by the tested criterion on the interval of point 2.
4. If after optimisation the best results do not show on the whole interval what is in item 1. - scrap the tested optimisation criterion for this TS.

The optimisation criterion is needed not for validation of robustness of the TS, but for finding the necessary settings of a robust TS.

[[Deleted]]

Aleksey Nikolayev:

The question was about other methods suggested by the author of the article.

In the article about books:

Ah well, I need to read more of his articles until I understand what he is getting at in the end

[Aleksey Nikolayev]

cemal:
Any plan to extend the article White reality? https://quant.stackexchange.com/questions/21163/whites-reality-check-for-pair-trading? utm_medium=organic&utm_source=google_rich_qa&utm_campaign=google_rich_qa

Not yet. I need to consider some ideas with one instrument.

[Aleksey Nikolayev]

Maxim Dmitrievsky:

Oh well, I need to read more of his articles, it's not clear what he's going for in the end yet

From my impressions - a loud headline and quite correct ideas. Partly similar to fxsaber's idea that the optimisation criterion should be consistent with the logic of the EA.

[Aleksey Nikolayev]

fxsaber:

Probably, this way of evaluating the adequacy of the optimisation criterion has a right to life:

1. We optimise, for example, by R^2. We get a nice straight line of equity.
2. We take half of that in point 1 as the testing interval.
3. Optimise by the tested criterion on the interval of point 2.
4. If after optimisation the best results do not show on the whole interval what is in item 1. - scrap the tested optimisation criterion for this TS.

The optimisation criterion is needed not for validation of robustness of the TS, but for finding the necessary settings of a robust TS.

I am not sure that it is necessary to optimise the optimisation criteria. It should be chosen depending on the problem to be solved. For example, as you suggested, it should be consistent with the system logic. Let me give you an example. If we exit at a fixed trailing stop, the distribution of returns will be close to exponential (with a shift). It will be defined by one parameter - the average. This parameter should be optimised for this method of exit, while it will be unsuitable for the other one.

[[Deleted]]

fxsaber:

Probably, this way of evaluating the adequacy of the optimisation criterion has a right to life:

1. We optimise, for example, by R^2. We get a nice straight line of equity.
2. We take half of that in point 1 as the testing interval.
3. Optimise by the tested criterion on the interval of point 2.
4. If after optimisation the best results do not show on the whole interval what is in item 1. - scrap the tested optimisation criterion for this TS.

The optimisation criterion is needed not for validation of robustness of the TS, but for finding the necessary settings of a robust TS.

have you tried to kill the sequence of trades so that the optimisation would depend less on them?

For example, randomly select the number of simultaneously opened trades at random intervals.

[fxsaber]

Maxim Dmitrievsky:

have you tried to kill the sequence of deals so that optimisation would depend less on them?

For example, randomly select the number of simultaneously opened trades at random intervals.

I don't get it at all.

[[Deleted]]

fxsaber:

On the one hand, it is assumed that the transactions are independent.

On the other hand, it is difficult to imagine a TS where transactions are independent.

For example, if you already have an open position, you cannot ignore this circumstance when deciding to make a trade.

It turns out that monte carlim is a non-random value. Or do I not understand something?

It was written here.

I meant that you can open several deals on signals (random number, different for each run in the optimiser). I.e. if 1 deal is already opened, then after a random time interval we can add another one, if there is still a signal, without waiting for the first one to close. I.e. it will allow us to make all trades independent from the previous ones in time... or pseudo-independent. It is relevant only when there are many trades following each other in close proximity.

Just a thought :)