The essence of optimisation - page 4

New comment
 
If the target function is known, then the optimization is trivial - change the arguments so as to obtain the desired value of the function. And if it is unknown? It was quite rightly said earlier that an example of such a target "unknown" function is random. It seems trivial and elementary but why are we talking about optimization on an unknown function - the market? This is impossible from the definition of optimisation.
 
There is a suspicion that the selection of a robust option (and without such a choice optimisation is not optimisation, this has already been dealt with) for an unknown process (and presumably not random), which is the market, can be done by approximation of a statistical set of parameters operating in different sections of the CB, following each other.
 
joo:
If the target function is known, the optimization is trivial - change the arguments in such a way as to obtain the desired value of the function. And if it is unknown? It was quite rightly said earlier that an example of such a target "unknown" function is random. It seems trivial and elementary but why are we talking about optimization on an unknown function - the market? It is impossible from the definition of optimization.

I think you are confused:)

"Target function" is not a model of the stochastic process itself, which would give an accurate prediction. It is simply one of the statistics like MO profitability. "Optimise" is to find an extremum of a function on a given target function, for certain arguments, with a minimum number of calculations, that is, to reduce the NP enumeration to quadratic or even linear with respect to the number of arguments.

In words it may be simple, but in reality the mathematics is not less complicated than that of those who master the new string theory, or as someone above said, analyses data coming from the LHC.

And this is not just to TRADE, but to make extremely obscene profits from this trade. Even a monkey can simply trade, and I am sure it can be trained.

 
m.butya:

I think you are confused :)

"Target function" is not a model of the stochastic process itself, which would give an accurate prediction. It is simply one of the statistics like MO profitability. "Optimise" is to find an extremum of a functional on a given target function, for certain arguments, with a minimum number of calculations, that is, to reduce the NP enumeration to quadratic or even linear with respect to the number of arguments.

In words it may be simple, but in reality the mathematics is not less complicated than that of those who master the new string theory, or as someone above said, analyses data coming from the LHC.

And this is not just to TRADE, but to make extremely obscene profits from this trade. Even a monkey can simply trade, I am sure it can be trained.


:О
 
I use Optimisation (enumeration) to find an opening/closing condition:
input int VarX = 0;
input int VarY = 0;

bool vars[];
vars[0] = MA1 > MA2;
// куча сгенерированных vars[]

if (vars[VarX] && vars[VarY]) OrderOpen();
After understanding this thing (well, that you can convert numbers into many things), finding a strategy became more interesting :)
 

Thank you gentlemen for so many interesting ideas and opinions!

Now I suggest shifting the conversation to a quantitative one.

Suppose there is a strategy with one parameter. Simple testing (1 - 1000) with 2 years of minute bars shows the following curve of profitability ( sum(PnL)/sum(abs(PnL)) )

That is, each point is the average MO for 2 years.

In your opinion, what points (parameter value) should be selected for prograding? For simplicity, you can directly indicate in the picture, with an arrow or by a circle.


Then I will tell you what conclusion I have come to.

 
toxic:

What locations do you think should be chosen for the pro-trade?

None, as there is no information about the risks of the strategy.
 
anonymous:
None, as there is no information about the risks of the strategy.

Yes, indeed, let it be Sharpe rather than MO, or MO/SCO then.

 
toxic:

Yes, indeed, let it be Sharpe rather than MO, or MO/SCO then.

If it is a Sharpe ratio - there is no point in selecting any point at all as there is a dominant portfolio.
 
toxic:

Then I'll tell you what I've concluded.

And what conclusion did you come to?
123456789101112
New comment