System Messages

This forum is for members of the Apiary Investment Fund. To learn more about how to get involved, sign up now or fill out the form below and preview the training right now!

MS ML Callout

 
Back to Forums
 
 
Mike S.

MS ML Callout

Legal Disclaimer: My Callout is for educational purposes only. It is not a recommendation to buy or sell. That is a personal decision that should be made by you only.

The ML callouts represent the evolution of my current trading skill. I don't claim they exist in a state of perfection or any given callout will be perfect.

 

Wed, 10/09/2019 - 1:17am
 
robunited

"I see partial correction in your review. But also a big view shift: "

No big shift at all ... just a complete run through of a system from A - Z.

Most people are stuck in their strategy only, and the high probability of it
... and I want to break down the trade management/risk management part of it ... separately.

"you depend on probability of risk more and not probability of direction to book your profit."

I see Strategy and risk separately.
Strategy/direction to book profit (regardless of how many pips that is in one run) and risk how to manage the losing trades.

We have discussed Setup mode & Strategy ... ok, done.

Now we come to Fixing Mode ... that is not about strategy anymore, that's about fixing the trades.

The truth is .. no one in the world, knows exactly what is going to happen.

A strategy is only as good as to provide some bias or direction.
Maybe some areas/entries of higher or lower risk.

But the initial positive R:R trade setup in itself ... is either a positive R:R, from the beginning, or not.

If it's not a positive R:R setup from the beginning, then the only other possible way is that
we are managing the losing trades during our trade process.
And that's what we break down now.

----------------------------------------------------------------------------------------------------

There is neither right or wrong ... both ways work.

Let's get back on track in the discussion, since we now have established that you are also
managing the trades, with an initial negative RRR setup (1:3), and you don't have an initial
positive R:R setup (as I initially thought).

That's an important point of clarification we have reached in our discussion, which was not
clear beforehand.

Next step ...

How are you managing the losing trades exactly?

"I explained that's not how I enter orders. I enter more like DCA
https://en.wikipedia.org/wiki/Dollar_cost_averaging"

Ok, that's the first time I hear this. I will need to read up on this, because I don't know how this works in detail.
Step by step Mike ;-)

------------------------------------------

One more thing, I have to compare this to the Wobble process, because you say you are Wobbling, and
our goal together is to see and compare which parts match/overlap, and which parts you add by yourself
and tailor it to your strategy.

There is no wrong or right .. just that run through of the steps and direct comparison.

Please don't feel it as a personal attack to your strategy ;-) You don't need to defend your strategy.
I believe in your regression lines as much as I do in any other form of indications, if this works for one.

I'm interested in basic structure, the break down of that complete system. Strategy & Risk/Trade Management.
As you write it down in a trading plan. And which parts are from the Wobble process.

The end goal is, to get a clear description of your strategy and how you manage the trades, and which parts are from the Wobble process and which parts are added and tailored.

This is a good exercise, because it shows and proofs how open the Wobble process is and everyone can take pieces out of it and make it their own. Which then work, for each one individually who went through the effort to make it their own.

I will read up on the DCA ...

 
robunited

So far, I think what we can conclude is:

The Wobble doesn't have a initial positive R:R trade setup, so doesn't yours.

The way you build up trades prior to the move, is based on the Wobble - in Setup Mode

But you use 1 trade per linear reg line, instead of layers at turning points.

Regardless how long we let the trades run, for 2 pip, 5 pip or 100 pip, at a certain time,
we need to manage the losing trades:

For the Wobble, the Fixing Mode tasks are clearly defined by Shawn (as listed above before).

Yours are different and based on DCA, which I need to read up on.

-------------------------

Another thing. The Money Management part of the Wobble is based on splitting up 1 lot into 100 pieces.
That's why you see so many little scalp trades in one bunch (i.e. 100 trades across multiple layers)

Do you split up your risk/trades in equal divided pieces or are you entering each trade as 'a new single trade?"
(Maybe that's the part of DCA which answers that question and I have to read up on first.)

In summary ... Strategy is clear, Setup Mode is clear ... Money Management & Fixing Mode not yet, but getting there ;-)

 
robunited

Sorry, I wanted to make this shorter or more to the point, but got distracted by a fire outbreak in a field behind the house ... everybody in panic, fire brigade, packing laptop and stuff together etc. ;-) It came pretty fast & close to the house, but now all good ...

Fire.jpg
 
Mike S.

Thanks for taking your time to analyze my strategy! You could charge a consultant fee for that!

"Do you split up your risk/trades in equal divided pieces or are you entering each trade as 'a new single trade?"
(Maybe that's the part of DCA which answers that question and I have to read up on first.)"

I'm not sure if I'm splitting them up. That would mean I identified an exact number of orders to only use in the first place. Since I have to adjust the regression lines every 24 hours, new lines might show up after the refresh that identify a price gap sufficient to put a limit order. But finding a new line only 20 pips or so better price is tiring, I prefer significant gaps like at least 50 to 100 pips. You're right I didn't calculate a way to identify the 'max' number of orders I'd use to hit a target. This though I think will only make underruse not overruse my risk order parameter (cause me to make less, not lose more). I only assumed I wouldn't use many orders since it's based on each new regression line at a significant price gap. And I assumed that entry technique (DCA) will compensate for earlier less profitable entries. I am interested to collaborate with you what my max order entry per session should be for my current account balance (I need a refresher here).

And I want to point out that the local min/max price action is fitting the multi regression line plots. I circled yesterday's local min/max points on the attached photo. My point is the price action fills the multi regression lines plots likewise it will fill the remaining lines that have not yet been touched. Those lines are all above the current price action, the highest in range being 1.091. The lines (targets) are dynamic, and I'll refresh the lines now and show that in the next comment.

2020-05-15 11_36_14-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png
 
Mike S.

Remember this analysis is linear (multi linear...)! Nothing to do with nonlinear analysis which is far superior resulting in +90% accuracy used in deep neural nets, my upcoming focus. But as you see it's still usable for some profit taking. Take care running from danger in the jungle there!

Attached is a refreshed chart. I wish I could refresh these lines real time using .mq5/4, well maybe soon...

The 1.092 target slipped a little, now is 1.090-1 range. I call this convergence to my current orders, just means I make less profit but because wobbling I should still make something each time. If you don't want to wait so long you can close the session at nearer target 1.087-8.

Lowest anticipated min next week 1.076-7.

2020-05-15 12_00_11-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png
 
robunited

"I'm not sure if I'm splitting them up. That would mean I identified an exact number of orders to only use in the first place. "

Identify and exact number of orders

That's exactly what Shawn is doing when he Wobbles in Setup Mode.

In Setup Mode ... he is building layers with equal number of trades, until a certain lot size has been reached.

Then he stops.

In Fixing Mode, he changes tasks/procedures/lot's.

"I am interested to collaborate with you what my max order entry per session should be for my current account balance (I need a refresher here)."

Allen has posted this video yesterday ... it opens up and can gives some hint's and clue's about RRR, S/L, Lot size for your account size ...

https://www.youtube.com/watch?v=k_Sf5ZcpwHM&feature=youtu.be

Also, I attach a quick screenshot, just in comparison of how Shawn would setup your trades, in terms of lot size.

All just food for thought ... and what we can take from it to make it our own,

Screen Shot 2020-05-17 at 7.10.26 AM.png
 
robunited

Now comes the important part of this. If you bring this all together, what's your true RRR?

You mentioned before ... "which is still less than shawn lucas' 1:12.5"

That would only be true, if you setup 1 trade full lot size, for 2 pip TP and 35 pip s/l.

But since that lot is broken down into 10 pieces ... in 0.10 lot size, and scaled in different layers, with imprecise entries at turning points.

(similar to how you scale out your 1 trade per linear reg)

What is the RRR now for 2 pip TP / 35 pip s/l?

Or what is the RRR then for your 100 pip TP / 400 pip s/l?

A true RRR can only be calculated exactly if you have a static Trade Setup, from the entry point, TP target and S/L.

If we do it differently, starting with a negative RRR, then it all depends on:

- How much do we put in Setup Mode only
- How do we fix the trades if it goes beyond (further against us then expected) ...

And what is the correct risk scale/balance of this, for our account balance?

 
Mike S.

Yes, Rob, that's what I need to talk to you about.

 
robunited

That's so very individual to what you have in mind ... so have to be very careful ;-)

But just purely from a risk control point of view, attached one example (example only)
how this could work for your trade setup.

You are in a good position, because you have a defined TP target
and defined entry points.

What you need to look for, I think, is a defined S/L which keeps you in Risk:Reward 1:1 or R:R 1:2
to be able to make an overal risk calculation.

And a process how you manage the losing trades before they hit S/L, for Break Even.

Meanwhile using these new available trades to build new positions at better places, the refreshed reg lines every 24h.

But always keeping your overall Risk in balance (whether this is 1, 2 or 3% risk) you want to put onto the table
... to find out whether your strategy was correct and you make profit.

If not, you should never lose more than your 'beforehand' calculated risk you are feeling comfortable to put in.
Anything can happen at any time, a instant move up or down 200+ pips, we have seen this before.

What Allen said, to get to higher levels in funding, Ron wants to see a smooth equity curve, not sharp cliffs up and down.

Setup Mode

That's the basic principle of the Wobble, building multiple layers, with always same lot size, to position
yourself at better places for the expected move.

The attached graph has more or less Setup Mode on top and Fixing Mode on bottom.

It's just a possible way. It's not easy to find the correct scale if we are managing the trades actively without
a set s/l and let it run and play out.

Maybe you can add an example which fits more what you had in mind.

I know that you want to let the trades run longer, without off-setting them.

Mike_One possible way.png
 
Mike S.

Interesting order management technique you have there for my targets and entries. I'll reply here more later.

BTW where did you learn to be so attentive to risk management? Lucas, Pyne, neither?

 
robunited

Watching and analyzing Mr. Lucas exclusively for the past 2.5 years.

Pyne sounds interesting from what Ed says ... also as he responded a bit on the forum. I will take a look sometime in future.

For the Wobble technique, basic principles, from Shawn only.

If you think about it ... you said before, you are not sure whether you are splitting up trades.

But, if you add multiple positions, at better places, you always split up, one way or another.

Otherwise you would just add i.e. 2% risk for the first reg line, another 2% risk to the second reg line, another 2% risk to the third reg line etc.

You have to get your overal risk for your trade setup first ... and then can decide whether to just enter at 1 entry point only, or build/scale this through multiple positions.

The question how you deal with the losing trades, which are running too far way ... that's something you have to elaborate further and see how it fits with your strategy/what you have in mind.

Also, you can test run this on demo accounts first. Don't make the mistake as I did, and learn and test everything on my funded ;-)

 
Mike S.

"You are in a good position, because you have a defined TP target
and defined entry points."

Thanks. It's good to know I have something valuable.

"What you need to look for, I think, is a defined S/L which keeps you in Risk:Reward 1:1 or R:R 1:2
to be able to make an overal risk calculation."

1:1 or 1:2? Quick question how does wobbling fit that RR?

"And a process how you manage the losing trades before they hit S/L, for Break Even."

I'm definitely interested in that.

"But always keeping your overall Risk in balance (whether this is 1, 2 or 3% risk) you want to put onto the table
... to find out whether your strategy was correct and you make profit."

The current way I enter orders the lowest I saw the 'Daily Return:' was -0.05%. This was a very low number and I'm impressed with it. I think in the future other scenarios will be similar. Likely will accommodate your 1,2,3% requirement easily.

"If not, you should never lose more than your 'beforehand' calculated risk you are feeling comfortable to put in."

What is this exactly?

"What Allen said, to get to higher levels in funding, Ron wants to see a smooth equity curve, not sharp cliffs up and down."

Thanks for letting me know, I didn't know that.

"It's just a possible way. It's not easy to find the correct scale if we are managing the trades actively without
a set s/l and let it run and play out."

Well you don't like my set stop loss idea (putting it clearly out of range of the current reg line trading channel). How would you define the stop loss?

"Watching and analyzing Mr. Lucas exclusively for the past 2.5 years."

Understood. Ya, the nature of machine learning is the maths are so good we focus on simply an answer we're looking for bypassing any theory on risk, unless that's what the ml problem is.

"If you think about it ... you said before, you are not sure whether you are splitting up trades.

But, if you add multiple positions, at better places, you always split up, one way or another."

Ya, you're right I just enter the orders. I operate of the belief that entering a 0.01 lot isn't going to cause much damage to a $2,500 funded account. And to a point I'm right; even with last week's total order drawdown at 1.0774 I was around -0.05% on 'Daily Return:'.
I hold the belief that a 0.01 order on a $2500 funded account will make $20 from 200 pips (per week), but at $250,000 the funded account will make $2000 from 200 pips (per week). I initially operated off that premise, I'm not willing to enter a new until a significant price gap like 50 or 100 pips occurs since I know exactly where the highest likelihood entries and targets are (thanks to the reg lines).

"You have to get your overal risk for your trade setup first ... and then can decide whether to just enter at 1 entry point only, or build/scale this through multiple positions."

Well, planning to use the max orders you can, just because the risk formula says that's how many you CAN use, I'm not convinced yet if that's what I SHOULD do. Like I said, why be greedy if $250,000 will make me $2000/week. The game is survival. You said risk is personal, if I don't feel I need to risk 5% of my account (by using that many orders) so I can have the peace of mind that in drawdown I'm merely -.05% loss... that's my style. But you're right, I need to know exactly what the numbers are, statistically, what's the average number of orders used per session, what's the average drawdown, how many orders does apiary risk say I can use, even though I don't feel comfortable using that many and will probably use a lot less.

"The question how you deal with the losing trades, which are running too far way ... that's something you have to elaborate further and see how it fits with your strategy/what you have in mind."

You have to remember you're dealing with a different strategy, not all strategies will need to douse the flames of loss as frantically as shawn's wobble technique. That's because he has no expectation to adjust his target higher than the original 2 pips and piling on up to 50 orders exponentially increases his risk/reward. He's quickly at the point of no return (per session) if he doesn't deal with his losing trades.

On the other hand as you see my targets and new entries take days, I enter a few orders (not 50). Targets are on the opposite side of the current price action range, so there's plenty of time to wait out price action, and watch the targets adjust. This is a target-centric strategy. I acknowledge risk, but I'm willing to use even less orders since I'm not focusing on risk to make money.

"The question how you deal with the losing trades, which are running too far way ... that's something you have to elaborate further and see how it fits with your strategy/what you have in mind."

Since I didn't plan to use as many orders as risk said I could, I imagined that a bigger stop loss than usual would be ok. If this is a perversion of risk rules, perhaps I should change it. I do plan to analyze this more to further minimize the stop loss and know how many orders I could use.

Attached current account health. In addition I'd like to add the alveo calculator rejects calculating any risk higher than 100 TP/SL. Is this a statement by Ron too? If yes, why 100 pips.

2020-05-17 19_36_33-Alveo.png
 
robunited

"1:1 or 1:2? Quick question how does wobbling fit that RR?"

Fairly easy. You have to step out of the assumption that the Wobble is only working with scalping for 2 pip in a one hour session.

That complete principle of building a position, managing trades ... can be applied to many different strategies, and over
a longer period of time.

Listen very carefully to what Shawn says, from 1.40min onwards, for just 5 minutes, in this video.

https://www.youtube.com/watch?v=Rsyy4fBKGvw&t=491s&pbjreload=10

And how this can apply to your strategy.

I think, so far, we have discussed a good bit about the Wobble principles and how they could apply to your strategy.

Maybe try to listen, or look at Shawn's sessions, from a different angle/point of view.

- With focus on how he structures/builds up positions in equal layers up to a certain lot size
- And how he starts to remove & manage losing trades

 
robunited

Back to your risk.

Yes, good that you only trade the min 0.01 lot and large gaps 50+ pips for new positions.

I also noticed that the max s/l is 100 pip in the OneClick setting. I don't know why that is.

"If not, you should never lose more than your 'beforehand' calculated risk you are feeling comfortable to put in."

"What is this exactly?"

I mean the Position Size Calculator, for your overall risk and lot size you can trade for your account balance.

https://www.myfxbook.com/en/forex-calculators/position-size

-----------

Alright, let's trade a little bit and please continue to post your trades, open trades, pending etc.

Let's see overtime which changes you will make to your trade management.

Screen Shot 2020-05-18 at 2.04.40 PM.png
 
robunited

In any case, that was great, I have a much better understanding now what and how you trade exactly ;-)

Thanks for your patience to go through it.

 
Mike S.

Ha! Ya Rob did you see that move last night? It took out all the reg lines in one swipe! I didn't expect my target to hit so fast. but it's a good example what I was referring to, the reg lines must hit.

2020-05-18 08_48_50-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png 2020-05-18 08_58_23-Alveo.png
 
Mike S.

Ok, so I woke up and all my targets were cleared. What's next? Back down of course. I'll adjust the lines and target down. If you zoom out, you see there's no targets above for quite a while (1.12). That's out of range.

2020-05-18 08_54_33-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,H1].png
 
Mike S.

So you can see it's making a nice sprawled out shape on the prior maxima target 1.09. There are literally no reg lines upward until 1.117 (1.12). So I think that's far out of range. A long term goal not relevant to this week. More relevant is the channel formed under the current price. A channel has formed between 1.09 and 1.08 since I'm at the max side of that channel I target the min side so my target is 1.082.

2020-05-18 09_48_44-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,H1].png 2020-05-18 09_52_16-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png 2020-05-18 10_05_00-Alveo.png
 
robunited

"Ha! Ya Rob did you see that move last night? It took out all the reg lines in one swipe! I didn't expect my target to hit so fast. but it's a good example what I was referring to, the reg lines must hit."

Yes, what a move yesterday! Great and really nice to see it hit all of your targets ;-)

Just a side note:

... nothing must hit, it's only probabilities.

The markets can be longer irrational than you can stay in the game.

Never forget that! Otherwise we will get toasted, sooner or later.

Hence, the risk management part of it, separate from the strategy, which creates consistency and a smooth equity curve.

Ok, looking forward for the next target 1.082 and the setup & trade management of it.

 
Mike S.

"... nothing must hit, it's only probabilities."

I was going to say that's because you focus on risk, but you said it for me the next sentence. If you're referring to 0.01% probability it doesn't hit, I can agree with that. The reg lines are dynamic remember, that's I can guarantee they hit.

You're too pessimistic, that's because you let people train you in technical analysis and not machine learning.

You know you focus on risk more than targets, too. I don't need to tell you. I appreciate your work in helping me understand the risk side, I look forward to work from you with this. But don't be too pessimistic about the TP side. You believe you have your risk managed perfectly? The target is nothing more than the inverse of the stop. How can you believe you can manage your stops perfectly but not your targets?

"The markets can be longer irrational than you can stay in the game."

The reg lines normalize (rationalize) the price action. Normalizing what appears irrational IS machine learning. You should get used to the idea that ML does this with 90+% accuracy. ;) To be honest TA trainers and ML developers are entirely opposite mindsets. I have no belief in traditional technical analysis tools or strategies, so naturally I don't believe in anything that any trainer says about using them for trading strategy including "...nothing must hit...", or "...nobody knows where the market is going..." This is fear mongering and has nothing to do with the use of high technology (vs low tech, which is retail technical analysis).

I can explain why the reg lines must hit at one point, which I did, but it's up to you to break out of the traditional technical analysis paradigm that nothing works very well (which it doesn't as long as you stay in that paradigm) and direction is irrational.

 
robunited

Mixing up pessimism with Risk Management is not a good thing Mike ;-)

"...nothing must hit...", or "...nobody knows where the market is going..." This is fear mongering"

With all due respect, this is not fear mongering. This is the plain truth of trading.

Risk Management is not fear mongering. It's key to trading successfully ... over a long period of time!
You don't have to believe me ;-)

Everyone is fully entitled to their opinion, of course. You won't change mine on this.

I have seen markets going against me ... forever. Too many times.

To say ... it must hit ... never. Not a single strategy in the world, whether its TA or ML.
Certainly not from a retail trading desk without any money behind to drive something.

I didn't want to put a damper on your optimism about your strategy, truly not, as you know, I like it and follow it ;-)

Just be careful with that.

 
Mike S.

When you start talking about machine learning strategies, you're starting to talk about very high probabilities. This is not 50/50 or 60/40 stuff. I understand your standpoint with risk.

"To say ... it must hit ... never. Not a single strategy in the world, whether its TA or ML."

The lines are dynamic. As long as time runs, they'll hit.

 
Mike S.

Maybe what one could think about instead, is why trading desks would prefer to brainwash people in thinking that moving average lines and indicators based on moving average lines are good for them to trade. That's clearly a misleading way to measure price action.

 
robunited

Ok, good ;-)

No need to argue or discuss this deeper, since everyone is trading with their own level of risk anyways.

It either works or not, over the long term. We all find this out for ourselves.

I just find it a dangerous way of thinking, that a strategies high probability is sufficient that it always 'must hit'.

 
Mike S.

"I just find it a dangerous way of thinking, that a strategies high probability is sufficient that it always 'must hit'."

That's why we have risk. But my point is how many times do you want your target to hit, 9 times? And how many times do you want your stop to hit, 1 time? So this is how much my mind is on each of them (9:1 target:risk).

 
robunited

"Maybe what one could think about instead, is why trading desks would prefer to brainwash people in thinking that moving average lines and indicators based on moving average lines are good for them to trade. That's clearly a misleading way to measure price action."

What do indicators, moving averages and price action have to do with Risk Management?

I'm talking about the way you measure and weight in the amount of money you are willing to put into something, and the level of loss you accept.

With everything in life, whether I buy a house, a car, investment into a Emerging Markets funds, Stocks, FX, Oil, Wheat or whatever.

Has zero to do with PA, Indicators, a ML strategy or anything else.

What does this has to do with brainwashing?

Just investments and the amount of risk you put it.

(brainwash ppl, fear mongering ... you start to sound like my cousin when he talks about his conspiracy theories on FB ;-)

 
Mike S.

Ok, I guess you're the risk teacher. I'm the student.

2020-05-18 21_15_00-Smith's ML Callout _ Apiary Fund.png
 
robunited

"That's why we have risk. But my point is how many times do you want your target to hit, 9 times? And how many times do you want your stop to hit, 1 time? So this is how much my mind is on each of them (9:1 target:risk)."

That's not risk. That's the probability of your strategy (winners vs losers) ... on a long run. You can either back or forward test this.

Risk is the amount of money you put in ... and how you manage it.

 
robunited

"Ok, I guess you're the risk teacher. I'm the student."

No, it's a discussion! ;-)

To me, you are mixing up Risk Management with the high probability of a strategy/trade setup.

But when Risk Management is seen as fear mongering, pessimism and brainwashing, then it doesn't make much
sense to discuss this further.

You have made up your mind about risk management itself.

And it's not just a discussion to go through and how to apply it, how much money/ lot's to put in and to manage the trades.

It's far from me to 'convert' you into something ;-) We are all free traders and only exchange our trading experiences.

 
robunited

And even if you have a strategy 9:1, winning 9 out of 10 times.

That one time in loss, without proper risk management, is enough to destroy your account in one go.
Or to get you deep enough into a hole you never get out off.

 
robunited

It's funny, because the Wobble is trade & risk management.

--> Not a strategy ... it's a mechanism, a model ... call it what you want. Which can be applied to other strategies.

But somehow, I get the feeling that you see risk management as a kind of an insult and conspiracy theory (brainwash, fear mongering)
to your high probability strategy, because it's based on machine learning and always must hit targets.

I really question this.

Like I mentioned before, keep this separated.

I'm talking to my cousin from Holland since 8 weeks in this corona stuff ... he is all 'brainwash', 'fear mongering', conspiracies. etc. Hardcore. And I get a flashback when I hear you throwing around in these words.
I don't want to argue the 'existence' of risk management.

I don't want to go down that route when just discussing a simple risk management model/mechanism for a trade setup.

I do want to discuss how it can be done, but not argue opinions about whether or not it's needed or pushed into the brainwash/pessimism/fear mongering category, that's just not right, and quite frankly, too much.

 
robunited

It's unbelievable.

There is not a single person on this entire forum I can talk about the principles of the Wobble technique, just the pure simple methods of entering, off-setting and managing the trades, in correct risk balance ...

.. without getting their knickers in a knot about their "Strategies"

Take the time to plot out possible risk management scenarios.

Instead getting the true picture/opinion ... risk is 'fear mongering', 'brainwashing', 'pessimism'.

Why didn't you tell me before?

You guys want to Wobble (-> I want to be acknowledged as the first Apiary member wobbling with a trend" ...)

And then discussing it in detail, to make it happen ... and then it's all a brainwash, fear mongering at the end.

What?

 
robunited

I don't know, maybe I've been too long in isolation and stayed at home and watched too much news.

Next you are going to tell me that FX trading is a hoax and only ML is the 'ONLY' way.

it was just a discussion between members (not student/teacher) to exchange experience.

But I don't let myself 'brainwash' into something so important as Risk Management, which takes
years for traders to find out, above their strategy.

Trading, if you have a 2.500$ account to try around and a monthly paycheck on top, yeah good.

Get real! Trade for real, make money for real, pay your expenses from money you earn with trading.

And then ... you know how much worth risk management is.

Trading is not a hobby, it's real!

In that sense ... sorry, have to leave the discussion ;-)

 
Mike S.

"I don't know, maybe I've been too long in isolation and stayed at home and watched too much news."

Yes, I think you've been in isolation too long. You're starting to sound like you're losing it. I'll ignore your rambling for now.

Anyway, my target for 1.082 is holding strong.

The 1.12 target is looking closer than I initially thought, the last time it hit that reg line was March 15.

2020-05-19 09_42_48-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M5].png 2020-05-19 09_51_17-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,H4].png
 
Mike S.

Save a doomsday hyperinflation scenario... I give you the new daily adjusted reg lines.

Note how 1.082 target doesn't move much despite approaching a new 100 pip grid limit order.

2020-05-20 09_19_06-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M5].png
 
Mike S.

At -100 pips drawdown risk was reported at only 'Daily Return: -0.25%' !

An account at $2500 generating $20/week at $250,000 generates $2,000/week.

Welcome to 'my' strategy. This is not scalping. You can call it 'trend wobbling'.

2020-05-20 14_51_31-Alveo.png
 
Mike S.

"Take the time to plot out possible risk management scenarios.
...
You guys want to Wobble (-> I want to be acknowledged as the first Apiary member wobbling with a trend" ...)
And then discussing it in detail, to make it happen ... and then it's all a brainwash, fear mongering at the end.
What?"

I didn't mean it that way. I think it's a misunderstanding. Of course I care about the risk scenarios. I need to understand it. I apologize for seeming insincere. I have a lack of detail and experience in the risk department. Sometimes we get caught up in our strategies. It's pride.

"Next you are going to tell me that FX trading is a hoax and only ML is the 'ONLY' way."

I'm a trader. I'm dealing in eurusd because I acknowledge it has the most liquidity I can find. That's why I'm using it. If I find something that has more liquidity of course a trader will go to that.

 
Mike S.

I inverted the color to see the lines more. On the left I circled where support zones were forming. There's two. I'm shooting for the lowest side of the second support edge. The diagonals left to right give a good view of the momentum there. If it's under that (1.093) it's higher probability price action to 1.082. You can see smaller one day diagonals at the top right forming showing interest to move down.

Since the lowest line on the support is still 1.082, it's still my TP.

I anticipate it'll pause a little after hitting 1.09 then resume to work slowly through 1.088-1.082 range until it finishes up past 1.082. Alternatively it could choose to punch straight through the 1.088-1.082 thick support range with a spike or single day momentum the way it recently did with the up target.

And now I'm at 0.21% not -0.25%! I am expecting 300 pips after the completed TP.

2020-05-21 09_33_37-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png 2020-05-21 09_58_09-Alveo.png 2020-05-21 10_04_37-Alveo.png
 
Mike S.

Using AWR (average weekly range) you can set it vs TP to expect the turning point to the TP. And then start trend wobbling around there-ish or half way to there... Because it could need to override the TP. The current eurusd AWR seems about 200 pips. So around 100 pips from TP could be good point to enter or at least observe for the extrema.

It's not easy to find premade AWR functions. Why? Everybody's focusing on something else. Good. I don't care and don't want everybody thinking like me. This is the point where I need to make my own.

Here's example of AWR function (MT4) https://www.forexfactory.com/attachment.php/2474432?attachmentid=2474432...

2020-05-21 10_37_00-Smith's ML Callout _ Apiary Fund.png
 
Mike S.

According to the AWR if 1.1 was the real max the anticipated min is 1.08 which is very close to my predicted min using reg lines of 1.082.

2020-05-21 11_09_56-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png
 
Mike S.

It's doing exactly as I predicted. High probability to keep down after 1.093 and take more time to work through thicker resistance starting at 1.09 to 1.082. I closed orders in order to avoid weekend blackout. I'll reopen GMT Monday. I think I'm getting the hang of it's drawdown. I can use AWR I described earlier to predict max range side vs TP. I'll show this more next week. Still working on planning the total risk management (thanks robunited).

2020-05-22 07_49_19-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png 2020-05-22 07_56_30-Alveo.png
 
Mike S.

Some brief Stats. I seem to be hanging around a $2 expectancy lately. ;)

This is because I use only some .01 lots.

2020-05-22 07_52_56-Statistics _ Apiary Fund.png 2020-05-22 07_53_37-Statistics _ Apiary Fund.png
 
Mike S.

After updating: a nice range from 1.092 to 1.082. As I left off last week, it's working through this thick support to the target 1.082. Diagonal multi reg lines indicate mid length up momentum was broken and short length down momentum is taking it's effect. According to AWR of 200 pips, and a max 1.1 (May 21), there is room for a min until 1.08.

I'd suggest initializing trend wobbling after 1.084. I'm expecting the min somewhere around 1.08. Last time the price extended nearly 100 pips beyond my estimated max. So, this is what I expect this time, and is the zone where I do my trend wobbling.

And I re-entered all positions after the weekend blackout.

2020-05-24 16_16_18-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png 2020-05-24 16_20_56-Alveo.png
 
Mike S.

The nearly 100 pip incline yesterday caused adjustments. The range flattened out causing more defined min at 1.084; the range is starting to slope up. ;)

The AWR max hasn't changed so technically I still plan for a min between 1.084 to 1.08. I do have to be conscious of the reg line at 1.115 though. In case it is intending to go that high, I'll be wobbling from here to there every 100 pips and see where the min target ultimately falls.

2020-05-26 07_49_01-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png 2020-05-26 08_02_49-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,H4].png
 
Mike S.

This trend wobbling I think is ideal for ranging time periods. But I have not measured it on pure directional time periods such as the ones attached. There are some very straight trend time periods lasting half year, etc. With hardly any retracement activity.

An ideal strategy will need to be able to utilize both pure sloping trends (with no retracements) lasting months or years, AND sideways PA.

I prefer to visualize 'trends in channels' not 'channels in trends' to solve this issue.

BTW this is why I prefer to deal with trends in the first place, rather than other types of strategies.

2020-05-26 08_09_56-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,H4].png 2020-05-26 08_12_12-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,H4].png
 
Mike S.

A closer look at the above charts will show though, that price did cross one of the multi reg lines, the W1 which is currently at 1.15.

2020-05-26 08_32_25-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,H12].png
 
Mike S.

And my definition of beautiful!:

2020-05-26 08_42_39-2100683976_ JAFX-Demo3 - Demo Account - [EURUSD,Monthly].png
 
Mike S.

I've been thinking about this. What would happen to multi reg lines in a pure diagonal trend slope with very small to no retracements? Most of the reg lines will converge and mirror the trend PA slope itself. Thus, the lines will still be accessible. Granted I could record some losses due to needing to follow the strategy of using the reg lines as targets, or maybe I'd opt to wait it out as I enter fewer orders for further away targets. I haven't yet experienced this.

2020-05-26 14_23_51-Smith's ML Callout _ Apiary Fund.png
 
Mike S.

So as you see this strategy's all about the inverse extrema. Am I at a max? I don't know, but if I am I'll target that min, etc... As you can see the reg lines are making a very nice slope incline shape on course to the adjusted D1 reg line at 1.1145. The adjusted min is a stable 1.084 which is only 20 pip adjustment from the original target after 113 pip PA incline.

2020-05-27 11_44_13-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M5].png
 
Mike S.

The question of risk is a seperate issue. It didn't serve me to re-enter after the weekend blackout at a new position as the 5% risk rule is recalculated to any new starting position. 5 orders of .01 lots and now breaching -1.5% isn't a recipe I'm looking for.

2020-05-27 12_12_43-Alveo.png
 
Mike S.

Though I think price action accurately uses many multi reg lines as support/resistance, it might become incompatible with AF's 5% risk.
I find time better spent exploring more python/data science/ML. There are various algos such as LGBM I already used to identify s/r injection.

As well as working on the neural net. I'll use TA as features.

 
Mike S.

Looks like PA 'did' want to go up to 1.12 this week! I was wrong! It's actually 1.1177 after being adjusted (D1 regression line). As the chart shows the min target is still a paltry 1.083 despite a princely 200 pip AWR uptrend completion for the week.

This strategy may not be riskless enough to move around in ApiaryFund's 5% risk rule.
Due to that I will not keep using this technique after the (adjusting) 1.084 target finishes.
It will be interesting to see how Alveo manages my current open orders if they violated the 5% risk rule.
This technique requires grid type scaling into a trend in order to 'pick a top/bottom'.
This technique is not as robust as modern deep learning or other ml performance.

A closer look at the H4 shows the last time the D1 regression used was 2 months ago (march), and before that 3 months (dec). The next reg line is the W1 at 1.15, it's about the same distance as the D1 was to my min target but I take the antagonist side again and think the D1 level will be touched briefly and PA will rush to the current target min 1.083. Either way the min is still quite stable as the original target when this session began was 1.082.

2020-05-28 21_39_13-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,M15].png 2020-05-28 21_49_55-Alveo.png 2020-05-28 21_54_58-21686771 - MetaQuotes-Demo_ Demo Account - Hedge - [EURUSD,H4].png
 
robunited

"This strategy may not be riskless enough to move around in ApiaryFund's 5% risk rule."

That's what I was trying to explain to you.
Are you running this on your live funded account or on a test/sim account?

"It will be interesting to see how Alveo manages my current open orders if they violated the 5% risk rule."

As soon as you hit 5%, all trades will get closed and your account is blocked from entering any new trades till next day.

 
Mike S.

The last 27 months have been a sharp downtrend for eurusd. Hopefully this or next week I can bail out of my current sell orders...and this multi reg line technique. That'll be it for me with that. Time to move on.

2020-05-31 22_15_27-CTSAWP_3_13 - Jupyter Notebook.png
 
danielfowler91

$2 is a positive expectancy !

 
robunited

"The last 27 months have been a sharp downtrend for eurusd"

There are hundreds of things and global events which are changing and driving price, every day.

10+ years ago, no one could believe the EUR/USD went up to 1.60, but it did. Today it's 1.11.
Where is it in 6 month, 1 year, in 3 years, in 5 years? No one knows.

"Hopefully this or next week I can bail out of my current sell orders...and this multi reg line technique.
That'll be it for me with that. Time to move on."

That would be a shame, because there is nothing wrong with your setup or strategy.

It just needs a mechanism where you control and manage your losses, in time.

Something which is repeatable and get's you into profit, over a longer series of trades. (hope and bailouts are not part of this).

Give it some thought. Finishing up your own strategy beats everything you will learn or search for in future, and later facing the same
issue again. There is no strategy 'alone', which always will be in profit.

Whether you believe it or not, it's always the risk management part of it, which makes it a fail or success.

Whatever you decide, I wish you all the best with it. But personally, I believe you will safe yourself a lot of time by sticking to your strategy and working it out, testing it sufficiently (on demo), until it's profitable at the end.

No one can take this away from you ...

This is your opening statement:

"This thread is a public record for Ron and the Risk/Funding Team at AF to understand my strategy."

The strategy is clear, but Ron is not interested in this so much by itself ... he's interested in how you manage risk (their funds).

That's where you have to put the focus, my suggestion to you. Just continue and finish it up! ;-)

 
Mike S.

Including an indi that at least tries to measure the calculus can resolve the stop loss and correct entry time. The calculus can tell you when is the best time to enter, but not where the price is likely to end up, and why the PA is conforming to such patterns, that's what the multi reg lines can do. Inversely, the multi reg lines can't tell you the best time to enter! That's what the calculus indi is for!

There is no magic here. Just using the right math, the right way! And if you don't use it the right way, you don't get the right result.

Plain and simple.

2020-06-03 10_36_26-2100683976_ JAFX-Demo3 - Demo Account - [EURUSD,M15].png
 
robunited

Great, love to see that you are continuing and make changes/improvements to your strategy.

"There is no magic here. Just using the right math, the right way! And if you don't use it the right way, you don't get the right result.
Plain and simple."

Good, let's continue to get it all working in the right way. I'm sure with the correct way of managing the losing trades, you will work it out.
(I will not interfere with it again, just follow and watch, don't worry;-)

Let's not lose sight on what the very basic objective is here. Managing other peoples funds (not our private money).

All we have to do is being in profit 3 month in a row and request an increase. To proof that we can manage risk, so investors have the confidence and give us their money to manage. (not only Apiary, you can do this with hundreds of other 'funded' trade programs).

A plain and simple objective.

 
Mike S.

What a surprise that two week trend up! But the min target has only changed 80 pips. Now at 1.09, was 1.082. Using an indi that can try to react to momentum helps me understand when to trade to that min target. So using that indi, instead of fading, I can just enter in my preferred direction and the prior extrema provides s/l.

2020-06-08 09_16_45-2100683976_ JAFX-Demo3 - Demo Account - [EURUSD,M15].png
 
Mike S.

I like your positive attitude, rob. Greatly appreciated.

2020-06-08 09_26_54-Smith's ML Callout _ Apiary Fund.png
 
Mike S.

#Correction: wrong data
"Before I say anything else, I first want to say that 1.3 is a stable target. I don't know how that fits in your 5% risk but in that view eurusd is oversold right now. "

For 2020 it is sloping down. (attached)

#But this is still right
Also I have seasonality data showing no matter middle of week, month, or year, the middle of such is the low. So v-shaped starting at beginning of the period to its end. That's a general analysis since 1971.

2020-06-16 10_47_51-fbprophet JCharisTech - Jupyter Notebook.png
 
Mike S.

The next 150 days.

2020-06-16 10_54_13-fbprophet JCharisTech - Jupyter Notebook.png
 
Mike S.

Next 24 days prediction using 200 data points (left).

Next 24 days prediction using 400 data points (right).

Not sure if I did it right. I'll look into it tonight. This is fbprophet, a nonlinear predictor. Say goodbye to multi linear regression!
https://facebook.github.io/prophet/

I'll use it later for minutes and hours.

2020-06-16 10_25_31-fbprophet JCharisTech - Jupyter Notebook.png 2020-06-16 10_27_52-fbprophet JCharisTech - Jupyter Notebook.png
 
Mike S.

1.2 looks like a good target today. Let's see.

Using stationary plot I can see PA is oversold and max range around stationary plot. (left)
Components also show that PA will maintain uptrend past July 1 and weekly lows typically are Monday and Thursday. (right)

If you want to use this you can git it at: https://github.com/gomlfx/apiaryfund/blob/master/fbprophet_%232.py

2020-06-16 23_07_41-fbprophet_#2 - Jupyter Notebook.png 2020-06-16 23_09_47-fbprophet_#2 - Jupyter Notebook.png
 
Mike S.

The above is daily data. I'll look at sub-daily data next.

 
Mike S.

The blue line is the predicted value. The black is the actual value. The actual isn't matching the predicted, but it does show an expected trend up.

Since the predicted line can be extended I'll do same for sub-daily data after I know how to import the timestamps.

Attached chart means I expect uptrend for the coming 2 weeks.

2020-06-17 16_26_24-fbprophet_#2 - Jupyter Notebook.png
 
Mike S.

If I look at something like this, I'd interpret it as possibly eurusd is in an uptrend, but the current price is too high. It should close closer to the prediction line that is at least 1.1-1.11. Even the predicted value 3 weeks out is not as high as the current price. This should be wobbling opportunity.

EOD (end-of-day) wobbling... so be wary of your 5% risk.

2020-06-17 16_52_19-fbprophet_#1 - Jupyter Notebook.png
 
Mike S.

Another point at the above. You can tell that the open values have been abnormally more volatile since 2020-3, the start of covid19 in USA. But this is summer isn't it? Well this is plague time now. Usually the gmt opens should be very smooth around the predicted value like 2018.

If PA = cash, somebody is making more than usual right now, regardless of (or because of?) the plague.

 
Mike S.

https://github.com/gomlfx/apiaryfund/blob/master/fbprophet_%231.py
This file uses MT4 .csv data files.

New daily opens continue lower headed to cross 1.1. You can download this code at the above link.

2020-06-18 19_56_58-fbprophet_#1 - Jupyter Notebook.png
 
Mike S.

I have the timestamp working. I did 1 day prediction using two weeks data.
That range shows monday expected downward continue of pattern to 1.115 with high at 12 noon GMT.
I think it says that because that's the pattern that's happened recently. But my point here is I seem to have the timestamp, so I'll run it on M1,H1,D1,W1,MN1, etc to get the big picture and that fbprophet is easy to use but claims to be very precise as well. So, there's more to come from this.

After I learn how to zoom in on the plot, I'll load more data while viewing the predicted region only.
Note the straight line in the middle of the graph is the weekend!

rough edition: https://github.com/gomlfx/apiaryfund/blob/master/June20fbprophet.py

2020-06-19 23_59_15-fbprophet_#2 - Jupyter Notebook.png 2020-06-19 23_59_52-fbprophet_#2 - Jupyter Notebook.png
 
Mike S.

uSING minute data. 1 day prediction, 20 days history. left chart:

5 min data 10 days history 1 day prediction, right chart

2020-06-20 08_58_22-June20fbprophet - Jupyter Notebook.png 2020-06-20 09_05_19-June20fbprophet - Jupyter Notebook.png
 
Mike S.

A lot of things to fix on this but, getting better.

 
Mike S.

For the week ahead:

Hourly data, 1 week prediction (blue line), trained on 1 year history (black dots). Given the other above predictions, I would say Monday will start down with a little upward highs to balance itself, then resume it's down momentum to fit the weekly prediction. 99% confidence interval (blue shade). I believe it will be a down week.

I have more adjusting to add with log scaling, seasonality, etc, so looking forward to this.

BTW heading to the ol' multi reg line 1.082 target quite nicely, eh? It all comes together...

2020-06-20 11_50_58-June20fbprophet - Jupyter Notebook.png
 
triguylm1

Mike,

This sounds solid. However, the probability of more upside is just as strong as European monetary policy has established some bullish sentiment. Figure price has to break below 1.1080 support to push bulls onto defensive mode.

Note the bullish trend since May 13 as price had tested 1.0750 support over 4 times in a month (April - May).

Granted price is running bearish since June 10. Which in my opinion is establishing a new price range from the high 1.1415 to this 1.1080 support.

Friday’s sell off has pushed price to set the stage for another bullish run in my opinion.

Anyway, use caution now as price will be testing support this week. I don’t know what the actual pivot price will be, but agree with price moving bearish first.

Thanks for your help with Euro analysis

 
Mike S.

triguylm1,

I plan to develop this a lot more. The focus is data science but I'll make eurusd the dataset. Thanks for your compliments!
BTW focus on intraday.

 
Mike S.

In Taproom today shawn scalped selling eurusd despite it being in a strong uptrend.

I also thought eurusd was down today, and wasn't expecting the attached.

Although I'll still work on directional prediction (this is science, not tea leaves) rob made the point that risk mgt can end you in profit even if your direction is completely wrong. Really, this is good news because even if my directional predictions are 100% wrong, if I am skilled at shawn's risk techniques I can still earn money. Ha.

Anyway, like he said being right at both is the best of both worlds.

Back to today's PA, it is not the intraday PA fbprophet predicted for Monday. I'll keep trying to improve the algo for intraday and observe the weekly and longer prediction ranges made earlier as well, which I used H1 and D1 data for. Intraday used M1.

To me, the intraday predictions are where things get exciting. Why? This is where things become more unpredictable likely because the biggest market players prefer that range (0-24 gmt) to trade against smaller traders. That would explain why yearly (monthly,weekly) trends still are very predictable but any single day is less.

2020-06-22 05_29_45-71086529_ MetaQuotes-Demo - Demo Account - [EURUSD,M5].png
 
robunited

Mike, if you want, or feel, to analyze the TOS session from today, it would be great to see your plotted trades for that session and how you traded together with Shawn. It provides a lot of insight. You can do it in the fade thread, if you want to keep the callouts on track.

 
Mike S.

Rob, I have a wobble thread. I should put it there. https://apiaryfund.com/forum/smiths-wobble-log?page=0#new

How do I plot the trades? That's still a mystery, I don't know how you do that. What do you think about private Zoom meetings for wobbling practice. Each zoom meeting is free 1 hour at a time last time I checked. Just renew the meeting after the 1 hour expire.

I came out flat that session because I was using most of the session to peel off loss orders. I think this is why I didn't make much profit, only 5 pips, since I was focused on balancing the account peeling off the oldest orders as soon as I booked any positive pips. It's great training for peeling though!

I think the great thing about wobbling is the profit isn't foundated on direction, but managing/balancing your realized profit vs unrealized loss. Therefore, theoretically you could wobble all day long. That is, if you know how to do it profitably.

 
Rookie

Mike See if these will help you plot the trades.

https://help.apiaryfund.com/how-do-i-plot-past-trades

https://apiaryfund.com/forum/how-use-plot-trades

 
Mike S.

Thanks, that's what I was looking for.

 
Rookie

NP Mike, glad it will help,

While the form search feature lacks functionality, I use google and just put Apiary Fund first in the search string works well.

 
Mike S.

H1 timeframe, weekly prediction: Fbprophet is fast to change it's directional prediction? Maybe it's used in real-time more. After the last two days up the prediction for the week shows an uptrend. Granted overbought just like the daily prediction says. I deleted seasonality accidentally. This is likely a rough sketch. It's important to get all the pieces right. I'll run M1 and D1 data next.

2020-06-23 09_09_07-June20fbprophet - Jupyter Notebook.png
 
Mike S.

D1 timeframe, monthly prediction: This might be a little wacko without the seasonality, but it says already oversold on D1 up trend.

2020-06-23 09_20_47-June20fbprophet - Jupyter Notebook.png
 
Mike S.

Since we don't focus on direction in wobbling, I wonder if it will be useful to make predictions for each our, since I will be focusing on intraday prediction anyway. And wobble sessions are usually 1 hour long.

Good prediction models rest on data points not time frames anyway, so it shouldn't matter which time range I use. The problem is having tic data or just M1. So I'll stay in M1 on 0-24GMT to test the models. This'll speed up testing the accuracy for each model.

M1 timeframe, 1 day prediction: downtrend prediction today. Let's stay sub-daily: hour to hour.

2020-06-23 09_54_00-.png
 
Mike S.

M1 timeframe, 1 hour prediction: down but note confidence interval. I used 1440 data points (1 day) to make the prediction for next 1 hour.

2020-06-23 10_18_38-June20fbprophet - Jupyter Notebook.png
 
Mike S.

Why new prediction on left didn't look fluid like old one on right... probably because I accidentally deleted some seasonality parameters. I'll make it better this weekend.

2020-06-26 09_50_31-MS ML Callout _ Apiary Fund.png 2020-06-26 09_59_53-MS ML Callout _ Apiary Fund.png
 
Mike S.

Zooming on prediction area now working. Trained on daily data 2012 to 2020 with 20 day (1 month) out prediction.

fbprophet indicating overbought and a return to normalcy around 1.09 for July.

2020-06-28 01_16_53-Python for Time Series Data Analysis Section10 Fbprophet daily data - Jupyter No.png
 
Mike S.

cross validation metric appears reliable.

2020-06-28 01_25_09-Python for Time Series Data Analysis Section10 Fbprophet daily data - Jupyter No.png
 
Mike S.

fbprophet 10 hours ago (left) vs updated just now (right).

H1 period, clearly not confident about the lower bound.

2020-06-28 21_11_33-fbprophet_non_daily_data - Jupyter Notebook.png 2020-06-28 21_15_49-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

1 week prediction range, H1. So I'm looking to hit 1.135 until the curve changes.

2020-06-28 21_21_05-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

As you see the predictions I'm getting for D1 and H1 data don't look nonlinear. So, I still have a lot of work to do on this. ;).

But I have learned that fbprophet nonlinear prediction is a step between deep neural nets and ARIMA models, so deep nets will be the next step after I master fbprophet.

 
Mike S.

1 day prediction range, M1. Not very nonlinear. What I know about these outputs is more linear = lower quality.

2020-06-28 22_34_52-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

Remember the daily period. After updating I think I still like prefer this frequency over subdaily.

Sub-daily data seems to suggest I should buy, but daily says price is way overbought and needs to head down.

2020-06-29 07_34_53-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

should be plenty of sub-daily info to time that daily normal, though. This H1 chart says no trend changepoint yet.

2020-06-29 18_37_58-fbprophet_non_daily_data - Jupyter Notebook.png 2020-06-29 19_27_18-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

This is what I want to know, which is right? I can test each hour. I have a feeling fbprophet autochanges from log to linear sometimes...

2020-06-29 20_05_01-fbprophet_non_daily_data - Jupyter Notebook.png 2020-06-29 20_14_07-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

Solved that I think. 'twas the seasonality problem.

2020-06-29 21_36_00-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

mind bending fun (and fund ;) ). Try 1.127 with a 1:1 R:R.

Tentative callout, but I won't be back for 7 hours.

Callouts using M1 data (two weeks), and 1 hour prediction.

 
Mike S.

1.127 hit but (my) prediction seems neutral atm at 1.1255.

I think I'll try to at least hit 1.126.

2020-07-01 20_25_42-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

Long term following up to 1.13.

Question is what can make that prediction curve change? fbprophet will let me know when that time came.

2020-07-01 20_24_49-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

Not bad. And still support for 1.13+

2020-07-01 21_41_07-Alveo.png 2020-07-01 22_13_36-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

So there was 1.13. One big drawback of fbprophet is it can't use gpu. Since dnn can use gpu and fbprophet is classed as less accurate than dnn's, it makes sense to replace fbprophet with a dnn prediction technique. It will be very nice indeed to see that day. Because I still need to buy gpu hardware :D

2020-07-01 22_47_16-Alveo.png
 
Mike S.

After the fall from 1.13 fbprophet is targeting a fast drop to 1.108-. I can use confidence interval for S/L.

What I learned about fbprophet is the prediction is a projection of what has happened on the recent time range. Almost like an extension of a current trend. No wonder why people say deep nets are smarter. But I am testing risk:reward on it to see if I can use it as a strategy.

2020-07-02 10_39_14-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

above call invalidated, prediction is flattening.

 
robunited

Many ways how to interpret this.

Might as well be that people took their profits yesterday and go into the long holiday weekend.

Will something happen today? Maybe, maybe not .... volume of trades, without the US, will
be down in any case today. On the flip side, low volume could also cause unusual volatile moves.

The Asian market never moves much ... let's wait for EUR/UK to open in 4 hours.

What's the statistic/history we can read on the charts, for all the 4th of July weekends with a smashing NFP report?

I guess my point is, every trading day is different on a micro level.

 
Mike S.

Flat to up. Note the widening confidence interval at the end of the prediction. This is uncertainty. No coincidence the weekend gap is 15 hours away.

2020-07-02 21_06_31-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

"What's the statistic/history we can read on the charts, for all the 4th of July weekends with a smashing NFP report?"

This is possible. I can do extract it asap. "...first Friday of every month at 8:30 a.m. EST. ..." is it?

 
Mike S.

Then flat. Better to follow this real time or shorter range.

2020-07-03 04_08_45-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

Like this.

2020-07-03 04_12_18-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

possilbe to scalp this stuff too.

2020-07-03 09_40_44-MS ML Callout _ Apiary Fund.png
 
Mike S.

I'll make sure to keep track of a 1 month, 1 week, 1 day, 1 hour range prediction starting next week.

I think 1 year history with H1, M15, or M5 res is workable.
fbprophet doesn't use gpu so higher periods is faster.

 
Mike S.

Here's top down prediction example:

using D1 since 2012 the next month shows overbought and should need to reach 1.09 this month.

using H1 (1 year history) the next week is oversold and should need to reach 1.14 next week.

I'll stop here. My point is to update that prediction daily and see how the curves change.

2020-07-04 00_47_46-fbprophet_non_daily_data - Jupyter Notebook.png 2020-07-04 00_58_37-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

rob,

"What's the statistic/history we can read on the charts, for all the 4th of July weekends with a smashing NFP report?"

I can extract a variety of stats what happened around the first friday of every month (8:30AM).

 
Mike S.

O does anybody think 1.13 just now was the high for today?

 
Mike S.

24 hour prediction using M15 dataset.

1 month training history.

2020-07-05 23_27_48-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

48 hours ;]

2020-07-05 23_32_26-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

The problem is how long it takes for me to run the prediction. Training two months M5 data was too slow. So I settle for M15, 2 months.

Voila, a 1 week prediction which I can update at any time.

2020-07-05 23_59_28-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

I can't help thinking the prediction is trying to show me this channel.

2020-07-06 00_55_00-71086529_ MetaQuotes-Demo - Demo Account - [EURUSD,M15].png
 
Mike S.

especially for weekly prediction, it still lacks the presence of where to put a stop and target. So maybe using apiarypivots in the direction of the curve could work for that.

2020-07-06 08_48_24-Alveo.png
 
Mike S.

Or I could stick to risk:reward a one hour prediction.

download (4).png
 
Mike S.

not bad, dnn is better...

2020-07-06 23_27_56-fbprophet_non_daily_data - Jupyter Notebook.png 2020-07-06 23_28_45-71086529_ MetaQuotes-Demo - Demo Account - [EURUSD,M15].png
 
Mike S.

weekly prediction changed from cyclic up to just up.

download.png
 
Mike S.

"weekly prediction changed from cyclic up to just up." Typo!

That was using H1 (1 yr). Here's M15 (4 months). It is good to contrast weekly prediction using various timeframes, though.

M15 looks undecided.

download (7).png
 
Mike S.

M15 (8 months). The more history I load them more it just hovers the prediction line above the current price.

Note the confidence interval (blue range) is narrower vs above M15 prediction w/ 2 months data. That is the key that the prediction has higher probability.

download (9).png
 
Mike S.

I have to remember, predictions are provisional. My current prediction is this week will take out yesterday's high.

 
Mike S.

M5 (2 months history). Wide confidence interval is what you don't want to believe. ;]

typo: confidence interval = uncertainty interval.

download (11).png
 
Mike S.

10:15AM Since fbprophet exploits seasonality (and I hate uncertainty) I decided to use 2+ years, instead of 1 year for H1 dataset to get a one week prediction.

Layering time cycles, such as years, months, etc is how fbprophet extracts predictions. So I think better to layer years.

10:19AM Finished fast but not impressive. Using more data always seems to produce lower sensitivity (flatten the prediction curves).

H1 (3 years) for 1 week prediction.

download (12).png download (13).png
 
Mike S.

I was right! And sooner than I thought. That's the power of nonlinear prediction! Next is deep neural nonlinear!

2020-07-08 10_09_54-MS ML Callout _ Apiary Fund.png 2020-07-08 10_11_06-2100683976_ JAFX-Demo3 - Demo Account - [EURUSD,H1].png
 
Mike S.

fbprophet's current sentiment for eurusd. To next wednesday is looking up.

download (1).png
 
Mike S.

provisional 5 hour prediction based on 80/20 train/test.

download.png
 
Mike S.

that provisional 5 hour prediction must had been detecting the retracement.

2020-07-09 08_09_39-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

Notice how the weekly uncertainty interval perfectly catches that current channel.

2020-07-09 08_17_16-2100683976_ JAFX-Demo3 - Demo Account - [EURUSD,H1].png
 
Mike S.

My first run on a (deep neural net) DNN and I quickly understood why I need GPU. Each epoch needs 1 hour on my non-gpu pc and here I'm using 25 epochs. That means I could only run one prediction on the weekend. Until I get a GPU. So I'll test H1 data for a weekly prediction.

This is also why people like LGBM (fast, accurate, w/out gpu) but I haven't found how to use that for time series.

2020-07-09 08_18_58-LSTM CTSAWP_Sec7_47_timeseries1 - Jupyter Notebook.png
 
Mike S.

Current trend component for M1 5 hour prediction.

I'm looking for a suitable intraday prediction range to use, maybe 8 hours.
Why 8 hours? For sleep, work, study intervals of course.

Or I might need to translate that to Tokyo, London, NY session. We'll see.

2020-07-09 08_30_03-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

Two days ago when I did the H1 prediction trained on 3 years history I got a flat line for the weekly prediction at 1.126. Now it's at 1.127 as I write this. That's no coincidence but the 1 year prediction still points to 1.14 for next week. Which is right? As we see now the answer is they both are! fbprophet is reasonably accurate with 1 or more years for the training dataset. I believe I can tell from the uncertainty interval if price is going to stay in the given range. If it's widens exponentially that's the give away it will not.

2020-07-09 19_17_30-MS ML Callout _ Apiary Fund.png 2020-07-09 19_24_27-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

btw the reason something like this might appear to miss it's prediction is because I am not at my pc at that time to update it... shorter prediction ranges using less training history etc. The 5 hour M1 predictions used I day training dataset.

2020-07-09 19_43_43-MS ML Callout _ Apiary Fund.png
 
Mike S.

Interesting how today's max was the inflection on the retracement at the above comment.

This is a good sign, meaning the predictions fbprophet is giving even at the above comment are significant.

I will experiment with it more and observe its ranges.

2020-07-10 09_52_42-2100683976_ JAFX-Demo3 - Demo Account - [EURUSD,H1].png
 
Mike S.

Widening uncertainty and flattening after 0 GMT July 14. I assume still some buying pressure on GMT Monday .

download (2).png
 
Mike S.

Trainers like to use daily data a lot when teaching their time series models. So I think I will stay here for now. But it is a small jump formatting to intraday data and I look forward to that.

I trained 1 year history for daily on fbprophet and was amazed how it looked like this weeks actual price action.

The trend component under it is another reliable tool.

2020-07-12 11_00_57-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

This week's prediction trend is accelerating up. Note the narrow uncertainty intervals.

2020-07-12 11_08_06-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

If I use close rather than open column I can get an update on the prediction at any time.

Including today's open (predicting open column) prediction shape attached. My point is update one time per day on a five day forecast out gives plenty of time to adapt to the price five days out.

2020-07-12 18_40_30-fbprophet_non_daily_data - Jupyter Notebook.png
 
Mike S.

If I let fbprophet auto choose the seasonality I get a much more modest target for 1.145 in 5 days (1 year training history on D1). I'd like to see which way is right this week.

2020-07-12 19_38_58-Clipboard.png
 
Mike S.

Note forcing seasonality (right) fits the dataset more. It should be fine for 1 week predictions. Linear is fine for linear algebra but not ok for the macro fractal ;). The added sensitivity should help me detect major turning points as well. [1 year training, H1, 5 day pred]

download (3).png download (4).png
 
Mike S.

Seasonality: H1 (left) vs D1 (right)

download (5).png download (7).png
 
Mike S.

I wonder if we'll get a high for today slightly over at 1.14.

If so I'd like to see if 1.13355 is the low today.

 
Mike S.

Not bad. How did I do it? I use an updated daily 5 day forecast and assume the prediction line is such a strong resistance that PA will max/min at it. Which is what happened yesterday. Since I knew the prediction line was at slightly higher than 1.14 yesterday I take the ADR and subtract to find the min 1.133.

2020-07-14 09_22_23-2100683976_ JAFX-Demo3 - Demo Account - [EURUSD,H1].png 2020-07-14 09_25_12-MS ML Callout _ Apiary Fund.png download (8).png
 
Mike S.

Today's forecast shows price (left) is already at the prediction line. So this is unknown territory for me. I am watching to see how it reacts. How it forms the min/max. But the prediction line direction appears accurate on train/test splits (right), even for seasonality. In seasonlity it will just exaggerate the final values but the direction will still be true for the nearest cross validation points (5 to 10). I only predict 5 to 10 days out anyway and update every day so that's not much a deal on a daily dataset, but I can see how that is more a deal on minute or hour periods where I try to predict 720 or even 24 data points.

today.png train test.png
 
Mike S.

hourly based seasonality for 1 week forecast, very flexible.

1 week forecast.png
 
Mike S.

Today's regular 5 day forecast. Looks like some concerns peaking out at 1.149 then heading down.

updated.png
 
Mike S.

The official open column based 5 day forecast. Somewhere in the background keeping an eye on that 1.146 TP 3 days out (not seasonal).

open 5 day.png
 
Mike S.

Using the hourly chart uncertainty interval to form a trading channel a possible channel trade is attached.

predi.png