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

Hi yoriz,

Tried downloading VirtualBox but it will not mount on my Mac. Also downloaded the earliest version of Mac OS needed to run MT5 on my computer
but unfortunately it is incompatible with updating from my current version of OS! Please can you reconsider uploading an MT4 version of your
'Fair Value Candles indicator'? I would be extremely grateful if this was possible.

Thank you.
Quoting ahmetdemir
prices do not follow the average, averages follow prices.
Yes. So in the cone-shaped graphs I made that showed the price distribution, I should also draw the MA closing in. For each of these “random walks” we can also calculate how the MA was updated in each step and when price crossed the MA.

With an EMA200 this is trivial, with SMA200 a bit more involved because the random walk bars must be preceded with the last actual 200 bars for the SMA to be correct.

From this we can calculate the expectancy: given a certain “spread”, what is the average profit/loss.

We use the same calculation to determine when to average down/scale-in: simply calculate if the expectancy improves, taking the improved average price but also the larger lotsize into account. Don’t average down until the stats show the expectancy improves by ‘x’ percent.
Quoting wingate
Please can you reconsider uploading an MT4 version of your 'Fair Value Candles indicator'?
Sorry to hear you have no means to run MT5. But no, I am not going to spend hours of my time so you can save money and keep using your 13 year old computer.

Quoting wingate
purchasing a newer Mac which due to financial constraints is not possible at the current time.
I hope you realize that to make money by trading you need capital to start with? If you can not afford to replace your old computer, I worry you can only trade using extreme leverage. Please be careful!
Quoting wingate
Hi yoriz, Tried downloading VirtualBox but it will not mount on my Mac. Also downloaded the earliest version of Mac OS needed to run MT5 on my computer but unfortunately it is incompatible with updating from my current version of OS! Please can you reconsider uploading an MT4 version of your 'Fair Value Candles indicator'? I would be extremely grateful if this was possible. Thank you.
Hi wingate,

Don't get blinded by wanting to use one specific indicator. If you are not able to run it, accept it and look for a strategy that fits your current situation and means.
There are so many good ways to trade even without any indicator. After all, every indicator is just another representation of price. Nothing more, nothing less.
It's never a holy grail, only a way to help us see things more clearly if we have the need.

Keep at it, this forum is full of great ways to trade.

The best of luck to you all!
Quoting yoriz
{quote} Excellent point! Indeed, I generated the stats based on the last 4 years which is a long time. Below is the corrected version using percentage return starting at each M1 bar of the last 4 years going forward up to 24 hours. Note that I added the 3 sigma lines also to add to Riskcuit's point about Kurtosis and fat tails. {image} Data was generated by the MT5 script below: {file}
Hi Yoriz,

In your percentage distribution graph, I think SquareRoot Of Time works perfect for future boundaries estimation.
I did a rough estimation from your graph. Here is how it goes:
From 1 hour percentage figure, the 24 hour estimation would be SquareRoot(24/1)*1 Hr percentage
From 6 hour percentage figure, the 24 hour estimation would be SquareRoot(24/6)*6 Hr percentage, etc etc.....

Very interesting..
Quoting CaesarCrypto
{quote} Hi wingate, Don't get blinded by wanting to use one specific indicator. If you are not able to run it, accept it and look for a strategy that fits your current situation and means. There are so many good ways to trade even without any indicator. After all, every indicator is just another representation of price. Nothing more, nothing less. It's never a holy grail, only a way to help us see things more clearly if we have the need. Keep at it, this forum is full of great ways to trade. The best of luck to you all!
to add to this, you don’t need a special indicator to see the candles. You can see them visually with just regular candles and a 200 MA, the custom candles are just the same data compressed into 4 data points per opportunity.

the purpose of them is just a visual aid to see what we are trying to do with this strategy. When trading though, you should use normal OHLC with the 200 ma.
ri · skuht
  • Post #207
  • Quote
  • Edited 11:15am Aug 27, 2023 10:57am | Edited 11:15am
Quoting GoldTheHun
I did a rough estimation from your graph. Here is how it goes: From 1 hour percentage figure, the 24 hour estimation would be SquareRoot(24/1)*1 Hr percentage From 6 hour percentage figure, the 24 hour estimation would be SquareRoot(24/6)*6 Hr percentage, etc etc.....
Good observation! This reminds me of this excellent thread by @AlphaOmega where he uses the same formula to model the price distribution over time throughout a single day:

Quoting alphaomega
This is why in the second system layer I use sqrt(t) to determine optimal entry points.
When I asked about the rationale behind the sqrt(t), I got the following reply from @StayFoolish:

Quoting StayFoolish
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.
  • Post #208
  • Quote
  • Edited 11:31am Aug 27, 2023 11:12am | Edited 11:31am
Quoting yoriz
{quote} Good observation! This reminds me of this excellent post by @AlphaOmega where he uses the same formula to model the price distribution over time: {quote} When I asked about the reasoning behind the sqrt(t), I got the following reply from @StayFoolish: {quote}
I’ll add to this as well, It’s not just price movements, it’s just built into any kind of process that has unique outcomes that are equally likely to occur. The more possible outcomes there are, the more unique pairs of outcomes there are (different sample paths).

If you compare the unique pairs of one level to the previous, the ratio infinitely converged to 4. An interesting side effect of this is, if you want to double the volatility of outcomes, you have to quadruple the length of the process.

this is why in any time series that is normally distributed, it takes 4 times as long for the series to move twice the distance.

Example:
- 2 possible outcomes = 2 unique pairs
- 4 possible outcomes = 12 unique pairs - 12/2 = 6
- 8 possible outcomes = 56 unique pairs - 56/12 = 4.67
- 16 possible outcomes = 240 unique pairs - 240/56 = 4.29
- 32 possible outcomes = 992 unique pairs - 992/240 = 4.13.

as you continue to double the possible outcomes, the ratio converges towards 4.

So if it takes on average 5 candles for price to move 20 pips, it will likely take 20 candles for price to move 40 pips (assuming it’s perfectly normal).
ri · skuht
Quoting yoriz
{quote} Good observation! This reminds me of this excellent thread by @AlphaOmega where he uses the same formula to model the price distribution over time throughout a single day: {quote} When I asked about the rationale behind the sqrt(t), I got the following reply from @StayFoolish: {quote}
I haven't read the book unfortunately. But from so many articles I have read square root of time is an excellent predictor for volatility.
For example to estimate weekly volatility from daily data, we would multiply the daily volatility by squareroot(7). Or 1 minute chart volatility to daily by multiplying it by squareroot of 1440 (number of minutes in a day). This rule will holds no matter what. I did so many tests few years back and the results are astounding...

Alphaomega's quote is very correct. I even started a thread here few years back
https://www.forexfactory.com/thread/...the-unbeatable?
Riscuit,

The method you recommend looks good except the averaging down part. Eventually averaging might get out of hand.
I am thinking about combining pairs so that they hover around certain mean and have a natural tendency to revert to the mean.
I haven't tested it, just thought about it. For example:
Long EURUSD
Long GBPJPY
Long AUDNZD

Short AUDUSD
Short EURJPY
Short GBPNZD

If you look carefully we are market neutral on all currencies but we have to find certain weights for each pair in order for this to work. We could equalize them by their volatility maybe, I honestly don't know at the moment.
But they should hover around around a certain mean in theory.
All ideas are welcomed
Quoting GoldTheHun
We could equalize them by their volatility maybe, I honestly don't know at the moment. But they should hover around around a certain mean in theory.
Have a look at this thread by @7bit where he does exactly that: make a linear combination of symbols to create a mean reverting synthetic symbol.
Quoting yoriz
When I asked about the rationale behind the sqrt(t), I got the following reply ...
The sum of N draws from a normal distribution with mean µ and variance σ˛ is N times the mean (N.µ) and its variance is N times the variance (N.σ˛).
The standard deviation, which is the square root of the variance, is √(N.σ˛) = σ√N.
No greed. No fear. Just maths.
Quoting GoldTheHun
Riscuit, The method you recommend looks good except the averaging down part. Eventually averaging might get out of hand. I am thinking about combining pairs so that they hover around certain mean and have a natural tendency to revert to the mean. I haven't tested it, just thought about it. For example: Long EURUSD Long GBPJPY Long AUDNZD Short AUDUSD Short EURJPY Short GBPNZD If you look carefully we are market neutral on all currencies but we have to find certain weights for each pair in order for this to work. We could equalize them by their volatility...
I really like the idea of this approach of using multiple pairs to create essentially a “delta neutral” portfolio so there is minimal to no directional exposure on any one currency. This way, you are removing that volatility from your portfolio returns and isolating what it is we are actually trading (excess kurtosis). This would create a much smoother equity curve and reduce risk.

this is exactly why I wanted to share the strategy. It has a lot to it already, but it is and never will be 100% complete so contributions like this, exploring different areas how we can optimize our results is just what this thread needs so thank you for sharing!

I am going to explore this further and will share what I find on it here.

this all being said, you are correct that you can not just average down indefinitely or else it is guaranteed that you will get over your skis risk wise and most likely blow up at some point.

this is why one of the core tenants is to stay small. Keeping this in mind will save you from 95% of blow outs.

there are also some strategies to handle this I will be sharing shortly, first will be “rolling” the trade to extend durations and reduce risk but again, the main thing is to stay small.

Second is to understand you will not have 100% winrate. There will be times where theta eats your whole profit opportunity in the big moves so that you lose money based on average entry once price returns to fair value. You can’t avoid this just like in business there are times where you having losing SKUs or spikes in cost that produce loses for some products or for periods of times. It’s never about any one trade or move. It’s about the big picture, letting CLT work for you.
ri · skuht
Quoting GoldTheHun
Riscuit, The method you recommend looks good except the averaging down part. Eventually averaging might get out of hand. I am thinking about combining pairs so that they hover around certain mean and have a natural tendency to revert to the mean. I haven't tested it, just thought about it. For example: Long EURUSD Long GBPJPY Long AUDNZD Short AUDUSD Short EURJPY Short GBPNZD If you look carefully we are market neutral on all currencies but we have to find certain weights for each pair in order for this to work. We could equalize them by their volatility...
Another thought on how to optimize this, when exploring ideas, don't let perfection be the enemy of good. Will likely be impossible to maintain a completely delta-neutral position but even a 10%-25% reduction in portfolio volatility would be tremendous over time.
ri · skuht
Quoting Riskcuit
{quote} Another thought on how to optimize this, when exploring ideas, don't let perfection be the enemy of good. Will likely be impossible to maintain a completely delta-neutral position but even a 10%-25% reduction in portfolio volatility would be tremendous over time.
I totally agree Riskcuit. That should be the initial goal.
Quoting yoriz
{quote} Have a look at this thread by @7bit where he does exactly that: make a linear combination of symbols to create a mean reverting synthetic symbol.
Thanks Yoriz, I will be checking that thread....
Quoting yoriz
{quote} Have a look at this thread by @7bit where he does exactly that: make a linear combination of symbols to create a mean reverting synthetic symbol.
I checked it Yoriz. That seems to be using some kind of linear regression I think or optimized values for weights. I would rather have a robust method for weights which are based on current volatility, not optimization.
Here is a rough idea:
First of all calculating each pairs volatility starting with LN(close1/close2), say in 5 minute timeframe. Than take least squares moving average of it for a week long, so that would be 1440 bars.
Now we have the volatility of all pairs that we are interested in. Convert the volatility to dollar amount using pip dollar values.
Total the whole thing and divide each pair by the total. Now we have a percentage weight for each pair.
After we get the percentage weights we should get the inverse, in other words highest weight should be the lowest weight so that our basket volatility would be reduced.
I think we should be getting somewhere
Quoting GoldTheHun
{quote} I checked it Yoriz. That seems to be using some kind of linear regression I think or optimized values for weights. I would rather have a robust method for weights which are based on current volatility, not optimization. Here is a rough idea: First of all calculating each pairs volatility starting with LN(close1/close2), say in 5 minute timeframe. Than take least squares moving average of it for a week long, so that would be 1440 bars. Now we have the volatility of all pairs that we are interested in. Convert the volatility to dollar amount...
I think this method is a great way to start the weighting approach.

Regarding the previous post mentioned AUD/USD, EUR/JPY, GBP/NZD as baskets, the challenge in this is assuming direction for each. There will be many instances where you can't be short all 3 pairs because of how the system works. What if on 2 pairs price is trading at a premium and on the one, the price is trading at a discount? Are you creating some synthetic price similar to the thread linked so you would trade all 3 short till they returned to a synthetic "fair value"? I could be misunderstanding your suggestion though so please correct me if that is the case.

I think what's more important is, that you want to find as large a basket as possible where the correlation between all of the pairs in it is all as close to 0 as possible. This way the drawdowns and run-ups won't be correlated so the curve should smooth out and you trade each pair using the fair value model independent from one another (letting CLT take control) vs. letting the direction you hold on one pair, influence if you would be long/short another in the basket.
ri · skuht
Quoting Riskcuit
{quote} I think this method is a great way to start the weighting approach. Regarding the previous post mentioned AUD/USD, EUR/JPY, GBP/NZD as baskets, the challenge in this is assuming direction for each. There will be many instances where you can't be short all 3 pairs because of how the system works. What if on 2 pairs price is trading at a premium and on the one, the price is trading at a discount? Are you creating some synthetic price similar to the thread linked so you would trade all 3 short till they returned to a synthetic "fair value"?...
Hi Riskcuit,

Long EURUSD
Long GBPJPY
Long AUDNZD

Short AUDUSD
Short EURJPY
Short GBPNZD
This combo is close to market neutral if we have correct weights. This is a basket index, 6 pairs in total. As you can realise all the currencies (not pairs) are neutral. we have 1 long and 1 short of each currency (again not pairs).
The logic would be create the index chart, than wait for it to go above or below certain MA, than enter the trade with the whole basket with correct weights. Once the basket index touches the MA, you exit all positions.
I hope I am clear

To make it clear here is the basket summary:
Index= (EURUSD*Weight)+(GBPJPY*Weight) + (AUDNZD*Weight) - (AUDUSD*Weight) - (EURJPY*Weight) - (GBPNZD*Weight)

Longs are plus, shorts are minus....
Quoting GoldTheHun
{quote} Hi Riskcuit, Long EURUSD Long GBPJPY Long AUDNZD Short AUDUSD Short EURJPY Short GBPNZD This combo is close to market neutral if we have correct weights. This is a basket index, 6 pairs in total. As you can realise all the currencies (not pairs) are neutral. we have 1 long and 1 short of each currency (again not pairs). The logic would be create the index chart, than wait for it to go above or below certain MA, than enter the trade with the whole basket with correct weights. Once the basket index touches the MA, you exit all positions. I...

ahh I see! Thank you for clarifying this.
ri · skuht
0 traders viewing now