Trader's self-deception: distrust of forwards. - page 13

 
Youri Tarshecki:

Let's say you've determined the minimum with a "montecarl". HOW LONG does your story have to be? HOW LONG should your story be? Until you show a real way of calculating the optimization cutoff - all said is just a way of SAMOUBMAN and lyricism on common knowledge topics. Equities, I've already said, don't save you from fitting. All your logic leads to only one solution - test on ALL history. Nevertheless, you do limit yourself to this. Why? If you don't need forwards - test on all available history and your "stat validity" will be maximal.

My way is simple, I compare the sum of forwards obtained on the same plot, but with different steps. The best result gives the best testing step. And what are you comparing with?

By no means, the type of eqvity is not a sufficient condition. It was shown earlier by me that a posh eqiti, can be the result of a fit. And your forward will miss such a fitted TS. A simple example: forward test with 1 year period this TS will pass, because the overall quality of equity of the strategy is quite high:


Even the semi-annual forward will pass. But this does not make it any sweeter, because the TS is a fitted one.

 
Youri Tarshecki:


My method is simple, I compare the sum of the forwards obtained on the same plot but in different steps. The best result gives the best testing step. Since I don't want abstract reasoning, but to maximise profits, I compare the profits that the forwards show. And what are you comparing with?

It's a manual, time-consuming adjustment, but it's an adjustment. The width of the backtest zone is wholesale, the target function is forward profit. I.e. by going through various options you find the best one on history, but how can you be sure that this periodicity (embedded in the chosen backtest window width) is a pattern and will make sense in the future?
 
Shit yeah take the glue of the forwards throughout history and analyse it.
 
омбинатор:
Hell yes take the glue of the forwards on the whole story and analyse it.

"Gluing the forwards together" is already more interesting. And how are you going to glue the forwards together? By hand? There is no known software that can do it automatically.

 
Youri Tarshecki:

By this logic, the best type of optimisation is in the last 20 years. Better yet, over 100. The nature of the price chart tends to change over time. Selecting the depth of history is a separate topic. But there should be plenty of forwards - that's for sure.

347
Avals 2015.07.28 12:12 RU
Youri Tarshecki:


My method is simple, I compare the sum of forwards obtained on the same plot but with different steps. The best result gives the best testing step. Since I'm not looking for abstract reasoning, but for maximum profit, I'm comparing the profit which the forwards show. And what are you comparing with?

Youri Tarshecki: It's also a manual, time-consuming, but very accurate calculation. The width of the backtest area is wholesale, the target function is forward profit. I.e. by going through different options you find the best one on history, but how can you be sure that this periodicity (embedded in the chosen backtest window width) is a pattern and will make sense in the future?
The TS itself should adjust to the changing market conditions. Constant overshooting of parameters will not solve the problem https://www.mql5.com/en/charts/3755939/eurusd-m5-e-global-trade
Chart EURUSD, M5, 2015.07.31 03:35 UTC, E-Global Trade & Finance Group, Inc., MetaTrader 4, Real
Chart EURUSD, M5, 2015.07.31 03:35 UTC, E-Global Trade & Finance Group, Inc., MetaTrader 4, Real
  • www.mql5.com
Symbol: EURUSD. Periodicity: M5. Broker: E-Global Trade & Finance Group, Inc.. Trading Platform: MetaTrader 4. Trading Mode: Real. Date: 2015.07.31 03:35 UTC.
 
Youri Tarshecki:

"Gluing the forwards together" is already more interesting. And how are you going to glue the forwards together? By hand? No software that can do this automatically is known.

I'm not going to :) I get the glue right away. Yes, it's much harder, yes, I have to write an engine, which emulates optimization, but I get immediate results, which I can almost instantly trust.
 
Комбинатор:
I'm not going to :) I get the gluing right away. Yes, it's much harder, yes, you have to write an engine that emulates optimization, but you get results you can trust almost immediately.
And what do you mean by "emulating" optimization? I.e. clicker just clicks on MT terminal, but then only pictures can be glued. Or optimising with a third-party program that duplicates the work of the optimiser?
 
Yousufkhodja Sultonov:
The TS itself must adapt to the changing nature of the market. Constant overshooting of parameters will not solve the problem https://www.mql5.com/en/charts/3755939/eurusd-m5-e-global-trade
Forward analysis is not parameter overshooting, it is a simulation of an unoptimised future, just designed to find out how well the system adapts.
 
Youri Tarshecki:
And what does "emulating" the optimization mean?

This means a mechanism which optimises the required EA parameters right during the backtest according to the principle you want.

As a result, the backtest itself becomes a forward gluing.

 
Комбинатор:

This means a mechanism which optimises the required EA parameters right during the backtest according to the principle you want.

As a result, the backtest itself becomes a forward gluing.

This sounds very incredible. Whatever backtest parameter works, it can't turn into a forward, especially several, because if you optimise just one parameter, it's not a forward.