Simple Mean Reversion | Page 30 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

A Simple Mean Reversion Strategy 392 replies

Pairs Trading: Reversion to the Mean 22 replies

Forward test of my new mean reversion strategy 11 replies

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

Cointegration, Synthetic hedges, mean reversion in R, Tech Thread 0 replies

Attachments: Simple Mean Reversion
Exit Attachments

Simple Mean Reversion

  • #581
  • Edited 5:00pm Oct 21, 2021 4:45pm | Edited 5:00pm
My understanding was that the levels are chosen based on the assumption that price does a random walk. With a random walk you expect the maximum excursion from the day open price to expand as sqrt(t), hence the indicator shown in post #7.

To verify that assunption, I analysed the EURUSD M5 bars of the last two years. Graphs use percentiles instead of averages to better filter out noise due to extreme outliers. Between yellow and green is 1 sigma, between red and blue is 2 sigma.

First I calculated where the current relative price was during the day:

Attached Image (click to enlarge)
Click to Enlarge Name: Relative actual price per time of day.png Size: 149 KB

As you can see, this does not look like sqrt(t) at all. Next I calculated the maximum excursion from the daily open price:

Attached Image (click to enlarge)
Click to Enlarge Name: Relative maximum excursion per time of day.png Size: 111 KB

Apparently the EURUSD is not likely to make a new high/low after 19:00 when the curve flattens, but again no sqrt(t) profile.

Can someone explain why sqrt(t) is used?
  • #582
  • Dec 25, 2021 5:56pm Dec 25, 2021 5:56pm
I think you will like this
GBPUSD
Three years: 2010, 2011, 2012
Initial lot size was 0.2
Attached Image(s) (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 20 KB
Click to Enlarge Name: БезымÑнный.png Size: 23 KB
  • #583
  • Dec 26, 2021 9:30am Dec 26, 2021 9:30am
GBPUSD
2010-2020
Initial lot size 0.1
Win rate is impressive
Attached Image(s) (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 20 KB
Click to Enlarge Name: БезымÑнный.png Size: 23 KB
  • #584
  • Dec 26, 2021 9:31am Dec 26, 2021 9:31am
Quoting Ihlas
GBPUSD, 2010-2020. Initial lot size 0.1. Win rate is impressive
Looking good, Ihlas! Did you use the original rules from post #1, or did you make improvements?
  • #585
  • Dec 26, 2021 10:15am Dec 26, 2021 10:15am
Quoting yoriz
{quote} Looking good, Ihlas! Did you use the original rules from post #1, or did you make improvements?
TBH I haven't read the first post fully. I just downloaded the EA from here. And tested few things what I believe is true. So no improvements But there are good posts that I have read here.
  • #586
  • Dec 26, 2021 12:31pm Dec 26, 2021 12:31pm
Quoting Ihlas
{quote} TBH I haven't read the first post fully. I just downloaded the EA from here. And tested few things what I believe is true. So no improvements But there are good posts that I have read here.
Can you please let us know from which post you downloaded the EA you used?
I would like to test it as well.
Thanks
  • #587
  • Dec 26, 2021 12:35pm Dec 26, 2021 12:35pm
Quoting Merka
{quote} Can you please let us know from which post you downloaded the EA you used? I would like to test it as well. Thanks
https://www.forexfactory.com/thread/...3#post10847673
  • #588
  • Dec 26, 2021 4:22pm Dec 26, 2021 4:22pm
Quoting Ihlas
{quote} https://www.forexfactory.com/thread/...3#post10847673
Thanks
  • #589
  • Jan 3, 2022 7:29am Jan 3, 2022 7:29am
Hey all,

Ever since I came across this thread I started thinking of ways to actually make profit in the market without using technical analysis--which, besides analyzing MAs speed/acceleration/convergence/divergence, feels like pure astrology to me-- and fundamental/sentiment analysis--which, at a retail level, feels very hard to get a complete grasp on.
I have actively been trying to decrypt alphaomega's messages in this thread on mean reversion. I personally tried to understand what he meant through those messages, overall I had a few ideas but nothing I really fully trust per my experience.

Before I delve into it, I'd like to say one thing; I don't feel like laying down cryptical messages on this forum is of any help to anyone. I fully understand how someone who has an edge on the market would not want to share it. But then, no need to spread mystical stuff that could mean everything and its opposite on such a forum. It's not particularly encouraging or helping anyone (which would be the purpose of the forum to begin with); as a matter of fact, stating plain truths like "price moves during the day and usually does not close at open level; only a professional like me is able to understand this" is quite pretentious as a statement to begin with and has no business in helping other members here.

  1. At first, we discussed about buying/selling when price exceeds a certain threshold (given by a price diffusion model proportional to t^1/2), betting that the price always reverts back to the mean. I feel like this assumption is not completely true; locally, there might be a mean reversion process happening, but on a bigger scale, one instrument's properties such as mean or deviation has no reason to stay stable.
  2. Then, we discussed about the "coastline" vs. daily return. This part is very interesting; realizing how much movement there is on a daily basis vs. how much the open-close difference usually stands at is eye-opening on the market mechanics. Alphaomega stated that his strategy is to capture a fraction of that massive pip movement in between the open and the close of a given daily candle; when asked if martingale/grid is involved in his system, he says that there is no grid in his system but then, a few posts later, he says "grid is the tool"...
  3. Finally, there has been discussion around the fact that the market balances to 1. In which way exactly? For example, if you went to buy 1 EURUSD, sell 1 GBPUSD and sell x EURGBP (depending on current quotes) you would end up with a neutral instrument minus spread; the fact that it's neutral would be temporary as quotes always change (so this instrument would end up drifting and being more exposed to a certain currency overtime). Is the strategy simply buying the whole 28 currencies betting that it's equivalent to a heavily ranging instrument, on which we can use strategies such as grids (that assume the underlying instrument won't trend in time)? That would involve paying so much spread that I feel like it's not the right path.
  4. Also, I might miss something: do we have statistical proof that buying on current lows and selling on current highs during the first half of the day (or whatever first n hours) and closing at EOD on all instruments is profitable?


What are your thoughts on this?

  • #590
  • Jan 3, 2022 10:49am Jan 3, 2022 10:49am
Hi marengo,
I coudn't dercypt alphaomega's messages in every detail either, but as an algo-trader, they gave me some valuable hints for my search for statistical edges.
I would like to comment two of your statements:

Quote
At first, we discussed about buying/selling when price exceeds a certain threshold (given by a price diffusion model proportional to t^1/2), betting that the price always reverts back to the mean. I feel like this assumption is not completely true; locally, there might be a mean reversion process happening, but on a bigger scale, one instrument's properties such as mean or deviation has no reason to stay stable

At least for Forex, my statistical evaluations and tests tell me, that mean reversion strategies mostly have a better edge than trend following ones. It seems, that the old economic rule "low prices attract buyers and high prices attract sellers" is still valid in the long run.

Quote
Also, I might miss something: do we have statistical proof that buying on current lows and selling on current highs during the first half of the day (or whatever first n hours) and closing at EOD on all instruments is profitable?

I coudn't find any edge until now in buying/selling extreme values in the first x hours of the day.
  • #591
  • Jan 3, 2022 3:17pm Jan 3, 2022 3:17pm
Quoting osfx
Hi marengo, I coudn't dercypt alphaomega's messages in every detail either, but as an algo-trader, they gave me some valuable hints for my search for statistical edges. I would like to comment two of your statements: {quote} At least for Forex, my statistical evaluations and tests tell me, that mean reversion strategies mostly have a better edge than trend following ones. It seems, that the old economic rule "low prices attract buyers and high prices attract sellers" is still valid in the long run. {quote} I coudn't find any edge until now in buying/selling...
I would also add that he said he was using multiple (at least 4) different strategies at the same time, on all instruments and all “directions”.
My guess is that such a heavy trading setup is such that drawdowns would be compensated by other wins but this is not necessarily the case; for this to be a true statement, drawdowns need to happen enough, below a threshold that is lower than what most of us expect there.
He said he gave description of his 1st system (sort of reversion to the mean thing) and hints on the others (“3 models” of the market being range, sub-trend and trend—which, to this day, are still not identifiable until they happen).

Or maybe this was all fake to begin with?
  • #592
  • Jan 3, 2022 11:56pm Jan 3, 2022 11:56pm
Quoting osfx
Hi marengo, I coudn't dercypt alphaomega's messages in every detail either, but as an algo-trader, they gave me some valuable hints for my search for statistical edges. I would like to comment two of your statements: {quote} At least for Forex, my statistical evaluations and tests tell me, that mean reversion strategies mostly have a better edge than trend following ones. It seems, that the old economic rule "low prices attract buyers and high prices attract sellers" is still valid in the long run. {quote} I coudn't find any edge until now in...
I'm sure you find other things. What if mean is dynamic?
  • #593
  • Edited 12:26pm Jan 8, 2022 12:09pm | Edited 12:26pm
GU 2010-2021
Initial lot size: 0.2
Yes, the DD is huge compared with initial deposit. But, yet again it's profitable. It will be incredible if the huge losses controlled.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 20 KB
Click to Enlarge Name: БезымÑнный.png Size: 23 KB
  • #594
  • Jan 8, 2022 1:04pm Jan 8, 2022 1:04pm
Quoting Ihlas
GU 2010-2021 Initial lot size: 0.2 Yes, the DD is huge compared with initial deposit. But, yet again it's profitable. It will be incredible if the huge losses controlled. {image} {image}
would you mind sharing the EA you used along with parameters?

cheers,
m10021
  • #595
  • Jan 9, 2022 10:24am Jan 9, 2022 10:24am
Hello! Greetings to the author.
You can share the .mql of the DRS indicator?

Tranks!
  • #596
  • Dec 28, 2022 10:52pm Dec 28, 2022 10:52pm
Quoting yoriz
My understanding was that the levels are chosen based on the assumption that price does a random walk. With a random walk you expect the maximum excursion from the day open price to expand as sqrt(t), hence the indicator shown in post #7. To verify that assunption, I analysed the EURUSD M5 bars of the last two years. Graphs use percentiles instead of averages to better filter out noise due to extreme outliers. Between yellow and green is 1 sigma, between red and blue is...

hi, I guess the sqrt(t) may come from the book "Ocean Theory" by Jim Sloman .
the book states that the change in price movement is porportional to sqrt of time.
  • #597
  • Dec 28, 2022 10:58pm Dec 28, 2022 10:58pm
I have done some data digging on EU from 2012 and found the High/Low price usually form at level 4/5 (using 10 ADR).
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 4 KB

I do think this system has its edge.
  • #598
  • Dec 30, 2022 9:57am Dec 30, 2022 9:57am
Quoting StayFoolish
I have done some data digging on EU from 2012 and found the High/Low price usually form at level 4/5 (using 10 ADR).{image} I do think this system has its edge.
Care to explain this graph please? Looks interesting.
cheers,
m10021
  • #599
  • Jan 4, 2023 7:36am Jan 4, 2023 7:36am
Quoting StayFoolish
I have done some data digging on EU from 2012 and found the High/Low price usually form at level 4/5 (using 10 ADR).
Thanks for the info. Based on post #108 these levels would be between 0.1 and 0.15 * ADR * sqrt(hour +1). Time for some experimentation...
  • #600
  • Jan 9, 2023 10:46am Jan 9, 2023 10:46am
Quoting yoriz
{quote} Thanks for the info. Based on post #108 these levels would be between 0.1 and 0.15 * ADR * sqrt(hour +1). Time for some experimentation...
should be 1/34 to 1/2 * ADR * sqrt(hr+1)
0 traders viewing now