Martingale principles

[NZEFILI]

@Abade69 No bro.

[NZEFILI]

@Abade69 I strongly believe you have less idea on how martingale works, and how you can properly manage your trades. If used correctly, with the right strategy or on a reversal point, this is one of the Best money management strategy.

So if you do not know how to properly manage your trades or use martingale, don't discourage others about it. Because everyone sticks to what works for them.

[Abade69]

@NZEFILI "Martingale: One of the best money management strategies", that's a good one. I guess some people only learn the hard way.

[NZEFILI]

@Abade69 what is your idea about martingale.

If you want to learn how it's been properly utilized, you can learn it.

Major things to consider while using the strategy is

1. Account size

2. Pip step

3. Multiplication factor. If you master this, you can thank me later

[Abade69]

@NZEFILI You can trust me, I have backtested hundreds of martingale-grid-cost averaging free and commercial strategies and I would never put money into any of them. They got a bad reputation for a good reason.

If you are willing to share your strategy logic I can tell you exactly why it's not such a good idea.

Watering down the x2 multiplier of the martingale doesn't fix the main issue, you are still over-leveraging.

[roar]

@Abade69 @NZEFILI to add to the discussion, grid strategies work fine as long as the market is mean reversing. Is it though, thats debatable.

I would say most markets are mean revesing most of the time, but never all the time. And this is why grid strategies break, and this is why their effectiveness is often exaggerated.

Its much easier to make a good-looking EA than a good EA. This is why most EAs found in the internet fall to the former category.

[NZEFILI]

@Abade69 can you recommend a good EA that you use

[Abade69]

@NZEFILI You could check the MQL5 market, but I don't recommend it. Most of the strategies are overfitted, use martingale or only work in "special" brokers that have no commission, spread or slippage.

you should learn to create your own strategies and build a portfolio of them.

What works?

1. Diversification (a bunch of strategies with little or no correlation working together).
2. Walk Forward Analysis (re-adapting your strategy to changing market conditions). No strategy in the world works perfectly for extended periods of time.
3. Money management that allows the use of compounding (risk adjusted to 1% or 2% of your current equity).

I don't use a fixed stop-loss but include a market exit when things go wrong (it's a negative skewed risk-reward), and re-adjust my strategies in a monthly basis. It's a lot of work but I don't have to worry about my account blowing up overnight.

[sktsec]

@Abade69 said in Martingale principles:

negative skewed risk-reward

I have some idea of negative skew, but how about negative skewed RR? Is it something we can measure?

[roar]

@Abade69 but if you use walk forward optimization, how can you get a reliable track record for the EA when the strategy keeps changing?
How can you know the expected average profit is positive?

[Abade69]

@sktsec I use signals from different indicators to trigger my entries, then add a TP that is equal to the current x1 ATR (10). Most of the times it will hit thanks to the current momentum or even the brownian motion of price itself.

If the TP is not hit then a market exit will be triggered when the indicator signal switches. Risk-reward is "negative skewed" because most of the time a loss is x2 or even x3 times the size of a win. It may sound like a bad idea, however the high win-rate will off-set this negative skew.

Some signals have better adaptability and faster response than others, I have a collection of 278 different signals.

[Abade69]

@roar That's a very good question. I have completely ditched the idea of "stability" in my trading strategies and/or "parameter range stability". I have also ditched the idea of "statistical significance" (the idea that you requiere X number of samples for any configuration to have any predictive value).

The matter of fact is that trading strategies don't have any "memory" of the market and cannot "learn", be "fed" or "trained" data to be consistently profitable in the future, it just doesn't work that way. I focus on increasing the degrees of freedom and make it adapt to the current market conditions as best as I can.

I know the strategy has predictive value because it is able pass an extended walk forward test with multiple steps (more than 10). I register the out of sample results and compare the relative effectiveness of each signal and I usually get the most adaptable and fastest signal for the most recent history.

I use 4 or 3 month in-sample and 1 month out-of-sample for the tests. The more you increase the in-sample window the worse it will perform out-of-sample. I use Profit Factor to pick the best performer per cycle, I don't mind if it's an "outlier".

Try it out! It works.

[roar]

@Abade69 maybe I will try it sometime, its just difficult to abandon the idea of statistical proof...

[Abade69]

@roar

Although, I was not referring to "statistical proof" (numbers and/or metrics used to validate a strategy), I was referring to the "statistical significance" that is used for a "training" period (A.K.A the number of trades that is supposedly needed to increase the predictive value of a strategy).

Walk forward analysis is perfect for getting that "statistical proof" for a strategy validation as long as you focus in the out-of-sample periods. However, you would need to abandon the idea of getting perpetual "stability" out of the strategy or the parameter ranges. It's best to imagine that you are testing the adaptability and consistency of the process itself.

[Abade69]

@roar Here is a visual example:

It's the same strategy, same data-set and same instrument. The only difference is the in-sample (training) period.

The first row shows the out of sample results for each month when using a whole year (more trades) for in-sample training. As you can notice it loses -29.22 in 10 months. When using 4 months for in-sample (less trades) it increases performance, even if the parameters that used are always different (inconsistent) for each month. Adapting to current market conditions is x100 times more important than the number of trades used for training.

I know this will be a useful insight for a lot of traders.

[roar]

@Abade69 to play devils advocate: the statistical metrics for walk-forward system tells how good the EA is fitting different strategies (parameter combinations) to sections of history data, right? There is less evidence that the latest strategy will work today and tomorrow

[Abade69]

@roar I like when people use critical thinking.

Of course there is no 100% guarantee that the selected parameters will perform well in the next set of "unseen" data (A.K.A the future). What is implied here is that there are far better chances of getting good out-of-sample/live results if we pick these parameters based on most recent data, rather than chosing a VERY large sample of trades, conditions and a larger time-span far away in the past that is no longer relevant.

Walk Forward Analysis simulates a rolling process of re-optimizations where you get a series of in-sample and out-of-sample cycles. When going live you are just doing one more of these re-optimization cycles, like the ones you do in the "testing" phase. Validation comes from the fact that MOST of your out-of-sample cycles were profitable, break-even or small losses. You can do 10, 24, 50, or 100 to gain such confidence.

It's not perfect, but it's much better than using a single stage optimization, a backtest with no out-of-sample or without backtesting at all.

[roar]

@Abade69 that makes sense in a world where history data from long ago indeed doesnt tell anything about the present. I think the jury is still out about that.

I could say, for example, that support and resistance levels have been a real thing in the past, and they also have significance in the present. If we lived in that world, it would be more attractive to find the S/R strategy inputs that work throughout the history.

Who knows, ultimately your bank account will tell if you were right. Do you have any live trading results on the WFA system?

[Abade69]

@roar Every trader I know relies on historical data one way or another. Backtesting, optimization, technical analysis and indicators use it, even if the trader claims to use "pure" price action, "logic", fundamentals or "projections", they all use something that is in the past to give them confidence or making trading decisions , that's just fact.

Furthermore, if there was no relation between price series we wouldn't be able to exploit their movement at all, could we? (referring to the efficient market hypothesis).

And yes, only out-of-sample / live results matter, but I am not willing to share as I don't want any unwanted attention, already shared a lot (may do a full disclosure later).

You can easily replicate my experiments with any strategies you've got. Re-adapt them every month with the most recent data and you'll see an increase of performance.

BTW There is a lot of crap on YouTube, don't trust it.

[AlphaOmega]

@Abade69 , roar
Thanks for the insight! very interesting and agreeable to me. I am using martingale with a MACH strategy. It only starts trading when the instrument I am trading becomes in an extreme position. I made it from a small 6 blocks algorithm roar wrote many moons ago here. Of cause it has grown to 274 blocks. I am not clever enough to put into words as what you have describe as your strategy but I always had a notion it is too weak to use only back test and find an expert good or bad. Also many experts work only short term and then fade. This is enough proof to me that no expert is suppose to work from the past into the future and eternity.. It needs to change as the market change.. Say weekly, monthly or yearly depending the pair that is traded. EURUSD for example has last week changed because it was to overbought.. Thanks for the insight