Finding The Most Robust Parameters Using Optimization | Page 3 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Backtesting Optimization parameters 0 replies

Optimized Parameters in Metatrader Testing? 3 replies

OrderSend parameters change? 3 replies

pivot points and MACD parameters 0 replies

Can parameters run on different periods in the same EA? 4 replies

Attachments: Finding The Most Robust Parameters Using Optimization
Exit Attachments

Finding The Most Robust Parameters Using Optimization

  • #41
  • Dec 23, 2021 6:53am Dec 23, 2021 6:53am
Quoting yoriz
{quote} I see. So you suggest to try to combine strategies in a portfolio that maximizes the Risk Adjusted Return found by MC simulation. I see how that forces the portfolio to be diversified, so it will probably include strategies with parameters that are far apart like the 3 red boxes @PipMeUp selected in his heatmap in post #33. Does that also imply robustness? Or does diversification in general already imply robustness? @FXEZ seems to think so. The only thing that worries...
Yes, it is computationally expensive but tractable with nowadays computers. If that is the problem, you can always use Genetic Algorithm and just examine a set of parameters that are promising. Note that the backtest algorithm in its basic form can be easily parallelized. For example with a 20-core CPU (and enough memory), a proper implementation of the backtesting algorithm in Python can accelerate the backtest 20 times faster.

/Matt
Trading is the hardest way to make easy money...
  • #42
  • Dec 23, 2021 10:27am Dec 23, 2021 10:27am
Quoting robots4me
I'm afraid you don't understand what OOS is
Thank you, I think I know what OOS is. Perhaps you have a different understanding how to use the OOS data? What is a live market today, can be an OOS backtest tomorrow.

Please have a look at the diagram in post #10. The green blocks named "Test" are your OOS parts. Let's pretend it is the year 2000. The year 2001 has not happened yet. We optimize, fine-tune, polish our strategy as much as we like using historical data upto 2000, but we do not look at the data from 2001 (because that year has not happened yet, remember). In fact, we make sure that data is not even loaded in your trading software!

Once we are confident we have selected the best strategies, we "go live". In reality we would now have to wait a few months to see the results of our strategy selection, but luckily we can step in our time-machine, download the 2001 data, and find out in minutes! The data we used for that is completely Out-of-Sample. In fact, it was not even installed on our system while designing and optimizing our strategy.

We repeat the same trick in small increments moving forward to towards today. The step size can vary. I use 2 month steps, but for example @AlgoTraderJo uses tiny 1 bar steps (see "Machine Learning with AlgoTraderJo"). If we like the equity curve our combined OOS tests gives us, we simply repeat the exact same process for the last step. However, in this case there is no time-machine. We can not yet download tomorrows data, so we actually have to trade live and wait for the results. Given sufficiently high quality tick data and a good simulation platform, the forward test results should be similar to the equity curve we got by backtesting on the OOS data.
  • #43
  • Dec 23, 2021 10:53am Dec 23, 2021 10:53am
Quoting yoriz
Perhaps you have a different understanding how to use the OOS data? What is a live market today, can be an OOS backtest tomorrow.
If today's data is used for backtesting tomorrow then it is not live data -- it is historical data. I don't know how one can spin it otherwise.

@yoriz -- I appreciate your patience and humoring my comments. You are a gentleman and a prince. Best of luck with your projects...
  • #44
  • Dec 23, 2021 2:15pm Dec 23, 2021 2:15pm
Nice statistical charts. Just remember, forex is non stationary and non linear in nature. Therefore, I only use two things in backtest: 1. Sharp Ratio. 2. Margin Level. So far my EA is working wonderful.
  • #45
  • Edited 2:32pm Dec 23, 2021 2:22pm | Edited 2:32pm
Quoting MathTrader7
Note that the backtest algorithm in its basic form can be easily parallelized.
Indeed! And so can the MC simulation.

The Walk Forward Analysis can be done more efficiently given enough memory/storage. If we can persist the internal state of the EA and indicators, and all open trades at the end of the test interval, we can continue from there for the next increment. We create a circular/rolling buffer that drops the oldest trades and adds fresh ones. That way, we avoid having to retest the entire lookback date range again and again. Only the few ticks that are new need to be simulated. Effectively we created "incremental" backtests.

I have even considered "incremental MCs" using randomized mapping tables, so I can efficiently drop old trades from the equity curves and add the new trades. However, that requires a lot of complex bookkeeping and a huge amount of memory/storage, so the effort is probably not worth the slight gain in speed.
  • #46
  • Jan 1, 2022 9:03am Jan 1, 2022 9:03am
Quoting MathTrader7
{quote} Yes, it is computationally expensive but tractable with nowadays computers. If that is the problem, you can always use Genetic Algorithm and just examine a set of parameters that are promising. Note that the backtest algorithm in its basic form can be easily parallelized. For example with a 20-core CPU (and enough memory), a proper implementation of the backtesting algorithm in Python can accelerate the backtest 20 times faster. /Matt
Hello MathTrader compliments of the season and i wish you the best of 2022 please i need your assistance ,i want to automate the fractals alert indicator based on its 4hr time frame signals, i trade the 4hrs chart and you wont believe it im making pips with the signal. The idea is to wait for the diamond to appear. and the alert issued that is all i will be very grateful if you can help me and make a new year 2022 my year. I use the attached indicator for my TP if the day direction is DOWNTREND price will be under the pivot level i trade both London and NEWYork sessions and i only trade the USDCHF pair.All the ea need to do is work with the signal of the diamond .Thanks
Attached Image (click to enlarge)
Click to Enlarge Name: MATHTRADER 7 FRACTALS.PNG Size: 39 KB
Attached File(s)
File Type: mq4 Dna-Pivot.mq4   31 KB | 278 downloads
Attached File(s)
File Type: mq4 Fractals alert.mq4   7 KB | 298 downloads
  • #47
  • Jan 1, 2022 9:12am Jan 1, 2022 9:12am
Quoting Slingshots1
{quote} Hello MathTrader compliments of the season and i wish you the best of 2022 please i need your assistance ,i want to automate the fractals alert indicator based on its 4hr time frame signals, i trade the 4hrs chart and you wont believe it im making pips with the signal. The idea is to wait for the diamond to appear. and the alert issued that is all i will be very grateful if you can help me and make a new year 2022 my year. I use the attached indicator for my TP if the day direction is DOWNTREND price will be under the pivot level i trade...
Hi,
Diamonds appear two candles after what you see on the screenshot. The fractals indicator is not working the way you think.

Happy New Year!
Matt
Trading is the hardest way to make easy money...
  • #48
  • Jan 1, 2022 11:38am Jan 1, 2022 11:38am
Quoting MathTrader7
{quote} Hi, Diamonds appear two candles after what you see on the screenshot. The fractals indicator is not working the way you think. Happy New Year! Matt
Hey! Matt you're super thanks for your speedy reply,and believe me it's my expected reply from you I know all about the fractal style infact you have just described my strategy without even telling you I would have shown you my live account with with lite finance as at December or hope you won't mind if I post in my everyday trade profit the ones for December 21 is off but starting from 3rd Jan22 I will post for you to see,I'm learning patience and I don't go for big profit what I do is wait for that diamond ,if it appears in New York time I normally enter on the second to the last candle of the day so it's the alert signal I'm watching if there's none I won't enter can the ea also work like that pls help me like I said I trade only 4hrs. Thanks Matt
  • #49
  • Jan 3, 2022 1:27am Jan 3, 2022 1:27am
I would like to suggest my way, unlike anything that offered here.
Instead of finding the best set of params for each asset/pair, I find a fixed set of params that suits every asset.
If it works for all, i.e all major pairs in forex, it is meant to be robust enough.
I know it may not give the optimized profit for each pair, but in return it will give me assurance that it is stable for many assets that behave different to each other.
In other words, I prefer robustness with all assets together than optimized profit per asset.
3-5 years of test period for this multi-asset is enough.
  • #50
  • Jan 4, 2022 10:57am Jan 4, 2022 10:57am
Quoting MT4-EA
Instead of finding the best set of params for each asset/pair, I find a fixed set of params that suits every asset. If it works for all, i.e all major pairs in forex, it is meant to be robust enough.
Thanks for the suggestion. I have heard this idea more often, however, I never understood this. Each pair has its own character. Some have large trends, others are more ranging with an occasional regime shift to a new mean, CHF has been regulated to a fixed price until 2015, etc. Each pair has different dynamics.

What is the rationale for developing one strategy that works for multiple pairs?
  • #51
  • Jan 5, 2022 5:33am Jan 5, 2022 5:33am
Quoting yoriz
{quote} What is the rationale for developing one strategy that works for multiple pairs?
Simply because it proves its strength in any condition in the forex market for all major economies.
Of course you can set a different set of params for each pair. It may produce a better result, but this is rich people's problems , for you - first find a strategy that works for all with the same set.
I personally would not use any other strategy which works only for part of the major pairs.
  • #52
  • Edited 2:38pm Jan 27, 2022 1:59am | Edited 2:38pm
Based on your excellent suggestions, I took the effort to wrap my EA in a bigger EA that implements built-in backtest capabilities. The EA will now internally backtest itself with all possible input parameters over the last few weeks and compile a portfolio of the best parameters and trade them all simultaneously. After a few days it repeats this. While trading it adds up all the strategies in the portfolio with equal weights to determine what the net position should be. I don't have a netting account, so I wrote a backend that maintains a net position by scaling out existing trades or opening new ones in FIFO order.

This gives hopeful results. Below is a backtest on EURUSD 2016-2021 using a portfolio of the top 150 settings (sorted by profit during a sliding backtest window of 90 days). It rebuilds the portfolio every 7 days:
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 30 KB


Net position throughout the years:
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 76 KB

Next steps I like to study and try out are:

  1. how stable is the portfolio (always the same 150 settings, or is the composition really changing over time)?
  2. what happens if we more coarse quantize the market position (e.g. -5 .. +5 instead of -150 ... +150) to save on spread/commission?
  3. implement Monte Carlo to build up a portfolio of inversely correlated strategies, rather than sorting on backtest results (suggestion by @MathTrader7)
  4. create an ensemble instead of a portfolio (suggestion by @FXEZ)
  5. use only backtest results of last 'x' trades, instead of last 'x' days (suggestion by @robots4me)

Still a lot of stuff to try out and code to write, but the rising equity curve gives the energy to keep trying.

Your hints and suggestions were very helpful. Thanks a lot!

I'll keep you posted on my progress. To be continued...

  • #53
  • Jan 27, 2022 6:10am Jan 27, 2022 6:10am
Quoting yoriz
Based on your excellent suggestions, I took the effort to wrap my EA in a bigger EA that implements built-in backtest capabilities. The EA will now internally backtest itself with all possible input parameters over the last few weeks and compile a portfolio of the best parameters and trade them all simultaneously... I'll keep you posted on my progress. To be continued...
You are on the right track! Congratulations!

By looking at the equity curve, I can say that if you perform a Monte Carlo sampling, due to long term drawdowns (specially in the beginning of the curve), it will show you a considerable probability of "Risk of Ruin". Which is not a bad thing at all! It shall give you an insight how to improve the portfolio to reduce this probability.

/Matt
Trading is the hardest way to make easy money...
  • #56
  • Edited 8:59pm Apr 4, 2022 8:23pm | Edited 8:59pm
My apologies for not making it clear as to why was I posting on your thread without making it explicit and ending up with an irrelevant intro or which I now construe as a vainful thing to do.

I wanted to share my experience of tapping hippo like price jumps or price surges hugging two sides of tiny pip ranges in a consolidation zone while trading equity indices.

I realized my error . Now I have set up a thread on it, explaining the whole system meant to help risk-averse retirees earn some income daily.

https://www.forexfactory.com/thread/...verse-retirees

The machine I am referring to in my thread is a robust transaction execution tool ,Scale Trader, which is able to execute high frequency style trades and capture tiny price surges . It is available to users of Interactive Brokers trading platform.

https://www.interactivebrokers.com/e...caletrader.php

On my FF thread I have shared actual records of my trading using the bot for sideways and consolidation zones entirely based on my experience in planning manually the non-directional strategy.

In other words, I have become a bot operator looking for locations to set it up and trigger entries with the help of signals provided by my brain.

Once the Algo is converted into a MT5 EA, I am willing to provide my inputs of how I manually identify 15 to 16 locations on a trading day.

My experience comes from trading index futures only. MT5 is used by futures brokers and not MT4 though both EAs may be developed.

It would be a challenge for coders to devise a fully automated EA codifying my experience into actionable entry rules to generate signals for entry.

I would share my experience only with coders who want to make it open source code as a learning tool for FF programming members.
Practice makes a person perfect
  • #57
  • Apr 4, 2022 10:15pm Apr 4, 2022 10:15pm
Quoting Ihlas
{quote} Harry Potter says that market doesn't change, but the scale. I'm trying to spot this baby snake
I am also exploring the same world.
https://www.forexfactory.com/thread/...2#post13955092
Practice makes a person perfect
  • #58
  • Apr 4, 2022 10:20pm Apr 4, 2022 10:20pm
Quoting MathTrader7
{quote} You are on the right track! Congratulations! By looking at the equity curve, I can say that if you perform a Monte Carlo sampling, due to long term drawdowns (specially in the beginning of the curve), it will show you a considerable probability of "Risk of Ruin". Which is not a bad thing at all! It shall give you an insight how to improve the portfolio to reduce this probability. /Matt
Can you please if tradiling drawdowns will help in containing the risk of ruin?
https://www.forexfactory.com/thread/...2#post13952912

Thank you.
Practice makes a person perfect
  • #59
  • Apr 4, 2022 10:24pm Apr 4, 2022 10:24pm
Quoting MT4-EA
{quote} Simply because it proves its strength in any condition in the forex market for all major economies. Of course you can set a different set of params for each pair. It may produce a better result, but this is rich people's problems , for you - first find a strategy that works for all with the same set. I personally would not use any other strategy which works only for part of the major pairs.
A semi-automated system and a simple non-directional strategy work in the case of most of the indices or any instruments whose two-way trading costs meet 1 pip or 1 tick of the symbol.

https://www.forexfactory.com/thread/...7#post13952167
Practice makes a person perfect
  • #60
  • Apr 4, 2022 10:32pm Apr 4, 2022 10:32pm
Quoting amsb
Nice statistical charts. Just remember, forex is non stationary and non linear in nature. Therefore, I only use two things in backtest: 1. Sharp Ratio. 2. Margin Level. So far my EA is working wonderful.
Capital efficiency or margin level management is the key to be a consistent trader.
I use an execution algo which can capture surges in a non-trending environment.
https://www.forexfactory.com/thread/...7#post13952167
Practice makes a person perfect
0 traders viewing now