Simple Mean Reversion | Page 25 | 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

  • #481
  • Jan 12, 2019 2:50am Jan 12, 2019 2:50am
Quoting alphaomega
{quote} Yes the grid is the tool. But not the operator! You need automated central control unit to manage these grids for 28 pairs. The control unit starts the gird at the right time, and stops at the right time. Otherwise the whole thing blows up on a regular basis. On average every 10-15 days the machine breaks. I suspect that the market makers initiate the "break switch" on pseudorandom intervals in order to prevent HFT machines like mine. BTW Look at the difference between these 2 charts. EURUSD vs GBPAUD What is the first thing you notice?...
Hi AO,

Amazing visualization. Are you able to share this?
  • #482
  • Jan 13, 2019 9:10am Jan 13, 2019 9:10am
Quoting carddard
{quote} Hi AO, Amazing visualization. Are you able to share this?
The indicator is only for experiments/data analysis. I had the idea and coded this quickly, so it's not perfect but it works.

Attached File(s)
File Type: ex4 Historical Price Distributions.ex4   11 KB | 1,401 downloads
  • #483
  • Jan 13, 2019 5:24pm Jan 13, 2019 5:24pm
Quoting alphaomega
{quote} The indicator is only for experiments/data analysis. I had the idea and coded this quickly, so it's not perfect but it works. {file}
Appreciate it AO!
  • #484
  • Jan 19, 2019 9:50am Jan 19, 2019 9:50am
Quoting alphaomega
{quote} Stop loss at level 6 or 7.
AO, does the stop loss level still apply after 12 in this case?
  • #485
  • Jan 20, 2019 10:12am Jan 20, 2019 10:12am
Wow, alphaomega, you have gone really deep into this rabbit hole. I admire your dedication! I just wonder what is your motivation for all of this, what is your end goal? What will you do when you reach your desired level of achievement in this endeavor?
  • #486
  • Jan 27, 2019 3:36am Jan 27, 2019 3:36am
Quoting alphaomega
{quote} The indicator is only for experiments/data analysis. I had the idea and coded this quickly, so it's not perfect but it works. {file}
Nice visualization of the historical high/low by time of day
  • #487
  • Feb 28, 2019 8:19am Feb 28, 2019 8:19am
Quoting alphaomega
{quote} The best way is to use fractals from different time frames! I simply apply the standard formula with 5 bars. You can apply this formula even down to the tick level. BTW Just look at the EURUSD for example. How much movement there is! The total length is enormous! Thousands and thousands of pips every week! {image}
https://www.forexfactory.com/attachm...1&d=1544533746

This blue and red point is fraktals from different tf?
  • #488
  • Mar 14, 2019 9:29pm Mar 14, 2019 9:29pm
Quoting alphaomega
This is the most important trading day of the year! (So far) Your success depends almost entirely on your ability(method) to handle days like this. So pay attention. {image}
Looks like we are conjoined twins brother. My coast line principle is also based on the screenshot you posted. Not sure where you are with this journey but after numerous twist and turns, I ended up adding a "shoulder" to my pacific coast highway (your top portion of the range in the screenshot) to keep "pushing" centrifugally outward of the curb until a turnout/vistapoint occurs to make a U turn to travel inland. Doing this shit on one pair/time series quantitatively seems doable in theory and logic but extremely hard when factoring all the market variabilities. Still not perfect but getting there. Some days are like driving 90 mph taking perfect curves and some days, it crashes straight in the ocean lol
Staying in my lane...
  • #489
  • Mar 15, 2019 3:14am Mar 15, 2019 3:14am
Quoting VEEFX
{quote} Looks like we are conjoined twins brother. My coast line principle is also based on the screenshot you posted. Not sure where you are with this journey but after numerous twist and turns, I ended up adding a "shoulder" to my pacific coast highway (your top portion of the range in the screenshot) to keep "pushing" centrifugally outward of the curb until a turnout/vistapoint occurs to make a U turn to travel inland. Doing this shit on one pair/time series quantitatively seems doable in theory and logic but extremely hard when factoring all...
I try to solve it by using the models described in this post.
Quantify wich model we are in, cut the wrong one's and let the right one's continue and make up for the losses.
So far so good, but i'm not sure if i'm there yet, need more confirmation.
Persistence will get it. Consistency will keep it.
  • #490
  • Mar 15, 2019 5:05am Mar 15, 2019 5:05am
Quoting alphaomega
THINK! PATTERNS! MARKET MODELS! TOTAL NUMBER OF VARIATIONS??? CYCLES! TIME DIVIDERS! RELATIONSHIPS BETWEEN TIME AND PRICE! PRICE OVER TIME! PRICE MOVING OVER TIME! (Not time over price!) TIME IS THE KEY! TIME DETERMINES PRICE! (Not the other way) NUMBER OF MARKET MODELS BASED ON FIXED TIME CYCLES? HOW MANY? 3 ! 1. HORIZONTAL RANGE (Many variations) 2. RANGE WITH DIRECTIONAL DRIFT (Many variations) 3. SMOOTH DIRECTIONAL TREND (Many variations) WHAT ARE THE PERCENTAGES FOR EACH MODEL TYPE? 1 - DOMINATES! 2 - APPEARS FROM TIME TO TIME. 3 - VERY RARE!...

Interesting way of classification of the market. Once a market is classified in a given cycle (i.e. daily cycle), a strategy for that model is activated for the day.

The challenge is how to predict the model that we are in during the day.

the following is a potential quantification algorithm that should be run at each bar in the day, in order to classify which dimension we are in:


1. T = total number of bars bars since day start (i.e. using 1m bars)
2. U = count how many breakouts to the upside
3. D = count how many breakouts to the downside
4. U% = 100 * U / T
5. D% = 100 * D / T
6. Identify the model as following:
Smooth uptrend if: D% < 1%, U% > 30%
Smooth downtrend if: U% < 1%, D% > 30%
Ranging uptrend if: U% > 1%, U% < 20%, D% < 1%
Ranging downtrend if: D% > 1%, D% < 20%, U% < 1%
Ranging horizontal market if: U% > 1%, U% < 20%, D% > 1%, D% < 20%

This algorithm can also be implemented on any market cycle (i.e. any candle, such as 8H, 4H, 1H, etc).

The next step after implemention the above detection algorithm, is to:
1. decide what pattern to trade (some of the above or all of the above)
2. write strategies for each pattern
3. find a way to switch between patterns during the day, if the market changed direction

let's implement this and see where it takes us.
  • #491
  • Mar 15, 2019 5:39am Mar 15, 2019 5:39am
Quoting alphadude
{quote} Interesting way of classification of the market...
Great stuff alpa!
This morning my eur.chf trade was cut off, because an model 3 uptrend was detected.
Sometimes the prediction is right sometimes not, but better be safe then sorry, the other trades make up for the loss and even made me some profit, so i'm happy the way it's going lately.
Persistence will get it. Consistency will keep it.
  • #492
  • Mar 15, 2019 6:48am Mar 15, 2019 6:48am
Quoting Chake
{quote} Great stuff alpa! This morning my eur.chf trade was cut off, because an model 3 uptrend was detected. Sometimes the prediction is right sometimes not, but better be safe then sorry, the other trades make up for the loss and even made me some profit, so i'm happy the way it's going lately.
hi Chake, nice stuff.

it could be either model 3 uptrend, or model 2 ranging uptrend. perhaps it is too early to say
Attached Image (click to enlarge)
Click to Enlarge Name: T2762_sc_20190315-124633.png Size: 45 KB
  • #493
  • Mar 15, 2019 7:03am Mar 15, 2019 7:03am
I start trading @ Frankfurt open and the trade got closed @ 11 am CET (white vertical line), so this was the picture I had at that time.

Attached Image (click to enlarge)
Click to Enlarge Name: Image1.png Size: 82 KB
Persistence will get it. Consistency will keep it.
  • #494
  • Mar 15, 2019 7:59am Mar 15, 2019 7:59am
Quoting Chake
I start trading @ Frankfurt open and the trade got closed @ 11 am CET (white vertical line), so this was the picture I had at that time. {image}
was it discretionary trade or automated? long or short?
  • #495
  • Mar 15, 2019 8:02am Mar 15, 2019 8:02am
automated, short
Persistence will get it. Consistency will keep it.
  • #496
  • Mar 15, 2019 10:28am Mar 15, 2019 10:28am
Quoting alphadude
{quote} Interesting way of classification of the market. Once a market is classified in a given cycle (i.e. daily cycle), a strategy for that model is activated for the day. The challenge is how to predict the model that we are in during the day. the following is a potential quantification algorithm that should be run at each bar in the day, in order to classify which dimension we are in: 1. T = total number of bars bars since day start (i.e. using 1m bars) 2. U = count how many breakouts to the upside 3. D = count how many breakouts to the downside...
the collective brainpower we have among us is all we need to crack the market! Interesting way to quantify I must say. I have slept over this shit for way to long to understand that no matter what we do, there will always be some unique or weird market condition that will trigger the wrong algo causing a instant drawdown in our account and efforts.

We have to think in terms of taking actions that accommodate all possible models/market conditions/patterns etc so here's where my head is....


Any of these three models require two actions... either divergence or reversion. We could do BOTH but I cannot find any way to make it work without time based filters. I consider price to be ALWAYS ranging or ALWAYS trending and I am a subscriber of this basic principle. I don't care what patterns or models price is making over time. The change in SCALE is what determines whether it is ranging or trending. Range expansion from the first tick of session open is trending.... scale expansion when range has developed and broken is also a trending state (trend continuation by definition is trending condition) in my opinion. But I have tried both Divergence only and reversion only and both cause drawdowns as expected when there are dislocations in price movement which obviously are unpredictable as to when they will occur to switch our algo from Diverge to Revert. That is the key to our problems in my opinion.... and not to predict or anticipate what phase the market is in or going into. The holy grail is the understand and devise a fantastic quantified model which will remain agnostic to changing market conditions and "ADAPT" to any predefined model shown in his screenshots. It is a bit scary how AO and I have taken two completely different routes (mine being Divergent and his being Reversion) and arrive to the same conclusions. It has been relatively easy to do this across 28 pairs as we have the tools like correlation, currency strength etc at our disposal. Doing this model on just one time series on one pair is VERY HARD.... I would say impossible but I am not there yet to make that claim. This is what I have been doing for last 3 months with NQ. one pair, one timeframe, 4 divergent algos. 4 reversion algos, 1 pullback/bounce and then apply brute force at each tick to pick up which algo picks and take the trade.... this way, I am not relying or anticipating which market model to go with. All I am doing is taking the right actions based on divergent or reversion setup and let price be my guiding principle.... all good in theory but very hard to maintain a consistent positive outcome. I am not giving up on this fascinating area of research. This CAN be cracked imo. Last night I wrote a 2 page post (which I lost in my thread) describing 3 more research projects that I am adding to my long term bucket list :-

EDIT: And if I may add, the holy grail is understanding the "Opportunity Cycle Theorem" that I have mentioned a 1000 times in my thread. It is the universal thread that connects everything in one place across all these unlimited (not just 3) models of market making patterns. You would be starring that charts for next 30 yrs if there is no concrete understanding of the underlying uniformity that exists in everything around us. It's the Pi or fibo or gann or elliott or chaos or fractal or whatever. It is the universal form of denomination that exists that we all overlook.
Staying in my lane...
  • #497
  • Mar 16, 2019 2:12pm Mar 16, 2019 2:12pm
Hi AO,

please allow to ask some questions:

Quoting alphaomega
If there is so much "excess" movement in comparison to the actual open to close change, isn't this the most obvious market inefficiency?
I basically means, that most prices are often returned during the lifetime of a day, doesn't it? This somehow reminds me on Eurusdd's "transience and recurrence" - no worries, I don't think that betting on the recurrence for a particular price value isn't you key to success. But maybe applying some ranging market strategies?!

I think this is the reason why you concluded here that ranges are more often than ranges with drift and far more often than smooth trends.

Quoting alphaomega
Add the fact that the 28 pairs balance out most of the time, during the first half of the day.
What do you mean with "balance out"? I could think about two explanations:

 

  1. The sum of the coastlines of the FX majors is almost constanct (if we compare two [almost] consecutive days) plus an error. Then balancing out could mean that if there is a very small coastline on some pairs but the coast lines of the other pairs almost reached their expected daily level, you can expect that the lagging pairs catch up since there only very few occassions of very big coast lines (the single points on the far right of PipMeUp's image).
  2. The three possible categories of days (as explained in your post linked above) will have reached a state so that the category is identifiable till noon. Then we could focus on trading applying the correct strategy for the corresponding category. (I think this opportunity isn't very likely.)


Is one of it correct? Which one?

Thanks,
AC

  • #498
  • Mar 16, 2019 9:10pm Mar 16, 2019 9:10pm
Quoting alphaomega
{quote} Now that you found part of the "secret", what is your next step? Come on, PipMeUp you are one of the smartest people in this forum. If you can't find the solution to the problem here, then what is left for the rest of the forum? Here is what I'm thinking. If there is so much "excess" movement in comparison to the actual open to close change, isn't this the most obvious market inefficiency? Add the fact that the 28 pairs balance out most of the time, during the first half of the day. And we have all the stars aligned right in front of us....
I have a theory here to share.... but please keep in mind, all my "theories" are completely based on what I have seen via screentime or via quantified research so I will share something interesting. Unfortunately my math and coding skills are not as great so hoping someone can prove me either right or wrong.

I am responding to this post but in general applies to all your posts here on this thread and primarily comment on "capture small percentage of this total excess length" below but please note that it applies to 1) one pair 2) one timeframe only i.e. a price vs time series 3) only one trade open at a time 4) only one microlot

  1. The "small percentage" has to be less than the total high/low range for a strategy to have long term merit. Anytime the profitable pipcount increases the high/low range (what I refer in my terminology as the OCRange), a)you got lucky and caught the trend b) market is throwing a fat tail distribution 'allowing' you to make a profit. The moment you exceed this range, market forces will double up against your will to make more.

Now, there is a catch in the above statement and none of you have discussed the fact that Range is a homogeneous organism that is every changing and is never fixed (or should never be considered a "fixed range"). In my definition, there is never a Fixed Ranging market even if you are able to draw a straight horizontal line. If it is not obvious why, market events can and will always happen unannounced breaking the range into shreds along with any "fixed" logic you put in place to enforce the range criterion. In my view, it will work but only upto the point when you decide to STOP trading after your goal is reached and with this statement also, there is a catch... how do you define a goal when the range is not there yet.
Gathering all these elements in a basket, shaking it up side down and sideways, the only technique I can come up with has to have the ingredients that does not depending on "market conditions". It has to be agnostic. It has to be neutral. It should have no effect on when you start or stop or where you enter or exit.

Staying in my lane...
  • #499
  • Mar 17, 2019 4:00am Mar 17, 2019 4:00am
Quoting VEEFX
{quote} I have a theory here to share.... but please keep in mind, all my "theories" are completely based on what I have seen via screentime or via quantified research so I will share something interesting. Unfortunately my math and coding skills are not as great so hoping someone can prove me either right or wrong. I am responding to this post but in general applies to all your posts here on this thread and primarily comment on "capture small percentage of this total excess length" below but please note that it applies to 1) one pair 2) one timeframe...
Have you tried to apply this heuristic principle in trading? looks legit..
  • #500
  • Mar 17, 2019 6:01am Mar 17, 2019 6:01am
Quoting alphadude
{quote} Interesting way of classification of the market. Once a market is classified in a given cycle (i.e. daily cycle), a strategy for that model is activated for the day. The challenge is how to predict the model that we are in during the day. the following is a potential quantification algorithm that should be run at each bar in the day, in order to classify which dimension we are in: 1. T = total number of bars bars since day start (i.e. using 1m bars) 2. U = count how many breakouts to the upside 3. D = count how many breakouts to the downside...
It's the right way to go about it. First need to ascertain the occurrence of the trades while looking back from : datetime to: DateTime
Find occurrences based on a threshold of xxxx pips using xxxxxBars . At least 1,000 pips are produced on a daily basis based on 10 pip size.
The machine moving between high and low of say 150 pips is actually producing 100 trades of 10 pips size while moving up and down.
We have to target at least 30% of it daily. 30 trades of 10 pips will be 300 pips positive and 150 pips negative. Nett gain on each day 150 pips.
Practice makes a person perfect
0 traders viewing now