Menu
Sister Sites

Options

Bookmark Thread

First Page First Unread Last Page Last Post

Print Thread

Similar Threads

Simple Mean Reversion 621 replies

M5 Mean Reversion Strategy 65 replies

mean reversion strategy 8 replies

Forward test of my new mean reversion strategy 11 replies

Synthetic hedges, cointegration, mean reversion and similar stuff 419 replies

Attachments: A Simple Mean Reversion Strategy
Exit Attachments

A Simple Mean Reversion Strategy

Quoting Merka
{quote} @RisKcuit Thanks for your long post where the content is about option theory. How do you translate that into normal forex trading and the discusions we were having since last week?
that’s a good question and I will explain when I get time to write another long post.

the reason I started with explaining it in option theory is because everyone believed and knows market makers make money.

they do so using the ideas in first post. Now that we understand how they make money using models, we can start to dive into the model of this strategy, find the parameters that are most likely to error, and how we can trade them.
ri · skuht
Quoting Riskcuit
{quote} that’s a good question and I will explain when I get time to write another long post. the reason I started with explaining it in option theory is because everyone believed and knows market makers make money. they do so using the ideas in first post. Now that we understand how they make money using models, we can start to dive into the model of this strategy, find the parameters that are most likely to error, and how we can trade them.
To confirm validity of this method, all you have to do is:

1. Create ea and optimized on at least 500 trades finding the best inputs
2. After finding best inputs, trade forward 500 trades with these inputs
3. Results forward trading will clearly show "Yey" or "Ney"
Quoting driven18
{quote} To confirm validity of this method, all you have to do is: 1. Create ea and optimized on at least 500 trades finding the best inputs 2. After finding best inputs, trade forward 500 trades with these inputs 3. Results forward trading will clearly show "Yey" or "Ney"
the method of optimizing is a bit flawed. You can always data mine perfect parameters for a sample. But that will result will only stick it all of the inputs stay the exact same in the forward test… which they won’t.


better to make a model, play with the parameters that are likely to change to see the effect it has.

understand the impacts these changes have, and then essentially trade the errors of those parameters.
ri · skuht
Quoting Riskcuit
{quote} the method of optimizing is a bit flawed. You can always data mine perfect parameters for a sample. But that will result will only stick it all of the inputs stay the exact same in the forward test… which they won’t. better to make a model, play with the parameters that are likely to change to see the effect it has. understand the impacts these changes have, and then essentially trade the errors of those parameters.
Good ideas but playing with parameters will only hinder and clout your ability to trade forward. I like to play with statistics because it is all you have. 500 trades of optimization should cover various market conditions. 500 trades forward will qualify or disqualify this method. I bet even after 100 trades forward, one will realize if it is a Go or no Go.
I think what he means is building a factor model and trade the mispricing opportunities. You surely can create a strategy based on the model, backtest it and generate lots of trade samples to analyze.
Quantitative Trader
Quoting Riskcuit
{quote} that’s a good question and I will explain when I get time to write another long post. the reason I started with explaining it in option theory is because everyone believed and knows market makers make money. they do so using the ideas in first post. Now that we understand how they make money using models, we can start to dive into the model of this strategy, find the parameters that are most likely to error, and how we can trade them.
Thanks @Riskcuit
I will wait for your next long message that I'm enjoying reading it
Cheers
Inspiring topic! Thank you @Riskcuit!

I wrote a simple Expert Adviser* following the basic rules from the first posts:

- SMA200
- RSI14
- Scale-in only if SMA200 breached entry price of last trade
- Close entire basket of trades when price crosses SMA200

On AUDCAD with a spread threshold of 60+ pips that gives the equity curve below (Monte Carlo analysis suggests a 18% risk-adjusted return per year):
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 29 KB


I am intrigued by post #135. Indeed, you could simulate thousands of random walks based on historical volatility from the current price forward. This gives a "cone" that is symmetric around the current price. From that we can calculate the corresponding SMA200 responses and get a probability distribution that helps to judge whether it is wise to scale-in or to wait. Definitely something I am going to experiment with...

*) No, I won't be sharing the EA here, because I have no energy at the moment for the usual questions after posting an EA on FF (e.g. how do I install, I am getting no trades, can you port to MT4, can you add this new feature, etc.). Sorry.
Quoting Zookeeper85
The indicator below plots the distance between the MA-200 of EURUSD and the MA-200 of GBPUSD.
Can you please elaborate why you used these MAs rather than the EURGBP MA-200 versus its current price? It surely looks more smooth, but does it compare to @OP’s spread?
Quoting yoriz
{quote} Can you please elaborate why you used these MAs rather than the EURGBP MA-200 versus its current price? It surely looks more smooth, but does it compare to @OP’s spread?

Hello Yoriz,
The indicator in question was developed by me some time ago with the intention of capitalizing on the positive correlation between the EURUSD and GBPUSD.
Given that these two pairs ideally exhibit a positive correlation, my concept was to execute a strategy involving buying one and selling the other when a significant deviation occurs.

In essence, this strategy can be equated to trading the EURGBP pair (Mean Reversion strategy).


P.S.: I attached the indicator below
Attached File(s)
File Type: ex4 distance_of_two.ex4   7 KB | 186 downloads
Quoting Zookeeper85
intention of capitalizing on the positive correlation between the EURUSD and GBPUSD. Given that these two pairs ideally exhibit a positive correlation, my concept was to execute a strategy involving buying one and selling the other when a significant deviation occurs.
Ah, I see. I just thought that perhaps you used a novel way of smoothing the "spread" using A/X - B/X instead of using A/B directly. Correlation trading is also an interesting topic, but not for this thread of course.

Thanks.
Quoting yoriz
{quote} Ah, I see. I just thought that perhaps you used a novel way of smoothing the "spread" using A/X - B/X instead of using A/B directly. Correlation trading is also an interesting topic, but not for this thread of course. Thanks.
I see your point! Even though the thread isn't directly about correlation, I thought it would be prudent to mention it.
Buying the EURUSD and selling the GBPUSD is essentially equivalent to buying the EURGBP vice versa.

For this reason, it's essentially a matter of measuring the distance from the EURGBP to its ma(x) value.

Here you can see the aforementioned indicator alongside an indicator that measures the distance to the moving average for current symbol (in this case EURGBP). They are pretty much identical.
Attached Image (click to enlarge)
Click to Enlarge Name: EURGBPH1.png Size: 50 KB
Quoting yoriz
Inspiring topic! Thank you @Riskcuit! I wrote a simple Expert Adviser* following the basic rules from the first posts: - SMA200 - RSI14 - Scale-in only if SMA200 breached entry price of last trade - Close entire basket of trades when price crosses SMA200 On AUDCAD with a spread threshold of 60+ pips that gives the equity curve below (Monte Carlo analysis suggests a 18% risk-adjusted return per year): {image} I am intrigued by post #135. Indeed, you could simulate thousands of random walks based on historical volatility from the current price forward....
I get that this is just a quick and simple EA with the basic rules. But if you would have traded this live and after 1-2 years you still don't have any profits, what makes you think you would have continued the strategy for another year and finally become profitable as the graph shows?
Quoting OneandOnly66
I get that this is just a quick and simple EA with the basic rules. But if you would have traded this live and after 1-2 years you still don't have any profits, what makes you think you would have continued the strategy for another year and finally become profitable as the graph shows?
Good question. I have two answers to that. First, you never trade 1 strategy/setting, you trade a whole portfolio. Other settings or other symbols might already have given profit earlier. Here are some other symbols with the same EA:

AUDCAD, 32+ pips spread, risk-adjusted return 88%/year (note the high deposit load!)
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 29 KB


EURAUD, 131+ pips spread, risk-adjusted return 12%/year
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 23 KB


AUDJPY, 104+ pips spread, risk-adjusted return 26.5%/year
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 22 KB


Admittedly, the strategy probably needs a different tuning during 2020-2021, and only started to shine the last part of the backtest.*

This brings me to the other answer: to do this properly, it is not wise to backtest like this. It is not hard to curve fit even a very bad strategy to show a nice curve. However, what were the chances you actually would have picked these settings a year ago? For example, see the "Myth of Averaging" thread for good examples of curve fitting random(!) strategies that still look good.

Instead, you should run a Walk Forward Optimization that dynamically determines the best settings over small date ranges and uses that for the next trades. I am currently rewriting the EA to do exactly that. I will post once I have results.

*) Yes, the backtest is only up to April. I never backtest until today's date and do all my tests and optimizations up to some "older" date horizon. Only right before I am actually going live will I run a last sanity check on the most recent data to make sure the strategy actually works and was not curve-fitted.
Quoting yoriz
{quote} Good question. I have two answers to that. First, you never trade 1 strategy/setting, you trade a whole portfolio. Other settings or other symbols might already have given profit earlier. Here are some other symbols with the same EA: AUDCAD, 32+ pips spread, risk-adjusted return 88%/year (note the high deposit load!){image} EURAUD, 131+ pips spread, risk-adjusted return 12%/year {image} AUDJPY, 104+ pips spread, risk-adjusted return 26.5%/year {image} Admittedly, the strategy probably needs a different tuning during 2020-2021, and only started...
You're pointing out several weaknesses with this strategy.
First off we have the high deposit load you have on one of the tests. Even if the curve fitted backtesting show it can be done, you will never know in the future. Even with a walk forward optimization. When trading larger capital you will never get that high leverage, thus getting even higher deposit loads when scaling up.

By using several pairs at the same time that are correlated you will crash and burn sooner than later. Of course you can pick pairs that are non correlated and get an overall better outcome.

By using any form of martingale you will have the same bad outcome. Martingale will never be safe, unless you use some sort of stop loss. But then it will probably not be profitable anyway.

By just using a 200 moving average as a mean reversion, you are missing important levels and micro/macro structures. The importance of fundamentals that usually can’t be programmed in a usable way. The lack of volume in forex that show hard traded levels and reversals. There are of course ways to get real volume for major pairs, but most retail traders don’t know how.

Most people on this forum use Mt4, which is a flawed platform to begin with. Both trade wise and backtesting wise. It’s popular just because it’s rather easy to learn programming on and there are lots of indicators and EAs already done.
Mt5 is much better, but still flawed in some ways.

What I see here is just a retail dream. Maybe you will make this into a profitable EA, just don’t quit your day job in the process.
The real edge that a discretionary trader has over a systematic trader is trade management. Most discretionary traders like to set and forget but that's a fatal error. They literally give up their true advantage. This strategy works if you know how to manage the trade properly. Entry is always an easy part. If you want to do automated trading, you should focus more on trade management, too. It's a layer that rare people know how to do it.

Here is an equity curve of a counter-trend strategy that I trade since April or May, 2020.
For motivation only.

EURUSD 2018 to July 2023.
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 24 KB
Quantitative Trader
reading the market like an open book
Attached Image (click to enlarge)
Click to Enlarge Name: EURUSDM1.png Size: 19 KB
Quoting OneandOnly66
First off we have the high deposit load you have on one of the tests.
Note that I tuned the lot sizing of the AUDCAD backtest such that it exactly matched the dotted-line which is the Monte Carlo risk-adjusted return scaled at max. 35% drawdown at the 95th-percentile (which is just some arbitrary personal risk limit I use). It resulted in using 0.0831 lots per 1k equity for each entry or scale-in opened which is quite high indeed.

When combining more strategies into a portfolio, the equity curve becomes smoother. This means we can crank up the lotsizes even more with the same risk levels. You are right that deposit load becomes a problem then.

Quoting OneandOnly66
When trading larger capital you will never get that high leverage, thus getting even higher deposit loads when scaling up.
Can you please elaborate what you mean? The brokers located on tropical islands and in tax havens offer leverage up to 1:2000, but I don't want to deposit my hard earned money there. The better regulated brokers only offer 1:30 to retail traders unless you can prove you are a professional trader (which they say you can prove by trading large volumes ;-)). So that suggests that trading larger capital will actually give access to more leverage, not less.

Quoting OneandOnly66
By just using a 200 moving average as a mean reversion, you are missing important levels and micro/macro structures.
Absolutely! I think we are all looking forward to the next post of Riskcuit that adds more seasoning to the basic recipe.

Quoting OneandOnly66
Most people on this forum use Mt4, which is a flawed platform to begin with. Both trade wise and backtesting wise. It’s popular just because it’s rather easy to learn programming on and there are lots of indicators and EAs already done. Mt5 is much better, but still flawed in some ways.
Not sure if this is relevant to this Mean Reversion thread, but yes, most trading platforms have their weaknesses. AmiBroker, Ninja, JForex, Zorro, etc. also have their problems. I have an Interactive Brokers account, but their trading platform feels like we are back in the eighties. No platform is perfect. As long as you know how it models the market and what it ignores, you can do useful backtests.

Quoting OneandOnly66
What I see here is just a retail dream. Maybe you will make this into a profitable EA, just don’t quit your day job in the process.
I don't think there are many professional traders here on ForexFactory. This forum is targeted to retail traders (and their dreams ;-)). Most of us here should keep their day job, and do Forex trading as an interesting side job.
Quoting yoriz
{quote} 35% drawdown at the 95th-percentile (which is just some arbitrary personal risk limit I use). It resulted in using 0.0831 lots per 1k equity for each entry or scale-in opened which is quite high indeed. When combining more strategies into a portfolio, the equity curve becomes smoother. This means we can crank up the lotsizes even more with the same risk levels. You are right that deposit load becomes a problem then.
Depending on your accounts leverage and your brokers stopout level, 35% drawdown can be ok or risky as hell.
Most people here are retail traders and have accounts between 1k and 5k. They cant open that many orders before they have run out of margin on a 1:30 account.
Then again most people here strive at getting a high leveraged account in a bucket shop that allows them to overtrade.

Quote
Can you please elaborate what you mean?
A couple of years ago the normal leverage was 1:500, new regulations made it harder for retail brokers to lure traders into getting to high leverage. Now the norm in Europe and Australia are 1:30.
As you say, if you can prove you are a professional trader you will get higher leverage. But the higher leverage only applies up to certain amounts, then it gets lower again. When trading large orders there will be other factors that come into play, but nothing most people here will ever experience.

Quote
Not sure if this is relevant to this Mean Reversion thread

Why I brought this up is that most people here will use EAs and do backtests on mt4, which will give false outcomes. Almost everybody in this thread are waiting for somebody to make an EA out of this strategy. And then the circus begins.
Quoting Riskcuit
This a long Saturday post to hopefully nudge you so you start changing how you think about trading. How to use a model and profit from errors... I have written about it many times in the past but never as clearly as I want to lay it out here and it's how to use a model in order to trade. Many people on Twitter will often post about "entry models" or "smart money" models but can't actually articulate what the model is. I am not saying they don't work, they might. But what I am saying is, they aren't clearly articulating what they are actually doing...
Taking the ideas from this post, let's extend them to this strategy...

We will start with a very simple base model with the following inputs:
- Price Distribution: We will assume a perfectly normal distribution for now. (0 Skew, Kurtosis of 3)
- Current Price: AUD/USD Current Close
- Fair Value: 200 MA
- Volatility: Starting with the standard deviation of 15000 AUD/USD candles, we use this to determine the volatility for a price distribution of a series that is 150,000 candles long.

So starting with our base inputs, our expectancy at various entry levels are as follows:

Sample 1:
Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot 2023-08-22 at 10.51.51 AM.png Size: 8 KB


Sample 2:
Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot 2023-08-22 at 11.13.36 AM.png Size: 9 KB


Sample 3:
Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot 2023-08-22 at 11.12.53 AM.png Size: 8 KB



You can see a pretty clear trend emerge, this system gets beaten down pretty badly in a true normal distribution. In this case, you'd pretty much want to trade the opposite.

But there are problems with taking this to the market in that way. The price is not normal.

All of the sample paths above assume a Kurtosis of 3. The problem is, AUD/USD 60-minute kurtosis can often be as high as 13. Skew is basically 0.

So in the model above, the input that is erroring when you apply it to actual market data is the kurtosis assumed by the price distribution.

Because of this, here are how the same volatility entry levels look on the market's fat-tailed distribution.

Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot 2023-08-22 at 11.21.22 AM.png Size: 6 KB



Think about why the error in kurtosis would produce this result... I'll post why it does this later but want to see if folks can think about it and figure it out.

To help, here is a visual aid of how this kind of kurtosis changes the distribution relative to a normal one (red line):

Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot 2023-08-22 at 11.22.47 AM.png Size: 17 KB



Also, think about what increases/decreases in volatility would potentially do to this model...
ri · skuht
Because higher volatility leads to fatter tail. The hit rate of higher volatility entry levels increases and brings more profits when the market reverts to its fair value.
Quantitative Trader
0 traders viewing now