Why Smart Traders Watch the First Hour (Initial Balance Guide) | Page 19 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

4 hour strategy (the nausbot - 4 hour trader) 25 replies

Why does 4 hour bar close 1 hour later than daily bar? 6 replies

average true range for 1 hour and 4 hour charts 2 replies

X-1102 1 hour , 4 hour and daily Tf trades 7 replies

Create a verticle line at certain hour for every certain hour 16 replies

Attachments: Why Smart Traders Watch the First Hour (Initial Balance Guide)
Exit Attachments

Why Smart Traders Watch the First Hour (Initial Balance Guide)

  • #361
  • Mar 11, 2026 2:25pm Mar 11, 2026 2:25pm
Seriously you two, take your 10 yo school yard shit away from here..
I follow this thread as I trade similar, so dont fuck it up with this crap...
Im only here for the Guzzoline...
  • #362
  • Edited 4:05pm Mar 11, 2026 2:28pm | Edited 4:05pm
Quoting amvt
{quote} And to be fair, you are right, there can't even be a contest when you think permutation tests take milliseconds. Oh no you corrected that to 4 microseconds. My bad G, you got more infrastructure than OpenAI
Well, try it. Take PnL array of 1000 elements, run 10k simulations. How long does it take on your machine?
You criticized the test results with great confidence, while being wrong in such a basic thing.
Didn't think it would go this far (4ms on average)
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.webp Size: 57 KB


Do you have any suggestions about IB strategy btw?
Air force
  • #363
  • Mar 11, 2026 4:09pm Mar 11, 2026 4:09pm
  •  amvt
  • Joined Feb 2025 | Status: Developer by Choice Trader by Heart | 202 Posts
Quoting W0lfram
{quote} Well, try it. Take PnL array of 1000 elements, run 10k simulations. How long does it take on your machine? You criticized the test results with great confidence, while being wrong in such a basic thing. Do you have any suggestions about IB strategy btw?
Look at us fighting like 2 little school girls and the teachers telling us to behave

Honestly, that's why I never wanted to give in but I had to because you said too much. But anyways, on MC you are right, it's trivial computation, but maybe you didn't notice it, I expected permutation tests. And that's what I recommended. Doing naive MC will give you results but nothing to take away from it. I criticized because I knew the results are wrong. Even with deterministic systems, getting a p-value of less than 0.0000 is extremely rare. And when we talk about a system that has tail-risk baked in, the results are even more suspicious. Both of which I mentioned in the original comment. I did not naively criticize it, there was evidence and logic to it.

I also recommend you Timothy Master's book called "Testing and Tuning Market Trading Systems". He said the same thing I expected that, traders usually don't do such tests because it takes a lot of time. But it's the most robust way to gain insight into statistical significance, which by the way was the whole point

I don't really know the strategy, and it would be foolish to recommend anything and futile to act on my suggestions for the strategy. But that's what I was getting at when I replied to eadwardcap, he didn't respond so I let it go.
— Developed by MMQ
  • #364
  • Mar 11, 2026 4:32pm Mar 11, 2026 4:32pm
Quoting amvt
{quote} Look at us fighting like 2 little school girls and the teachers telling us to behave Honestly, that's why I never wanted to give in but I had to because you said too much. But anyways, on MC you are right, it's trivial computation, but maybe you didn't notice it, I expected permutation tests. And that's what I recommended. Doing naive MC will give you results but nothing to take away from it. I criticized because I knew the results are wrong. Even with deterministic systems, getting a p-value of less than 0.0000 is extremely rare....
Quote
I expected permutation tests. And that's what I recommended.
I get it, but we don't have any params here; if IB_bull -> buy kind of trading logic. If it was ma cross strategy, then we should have to vary ma window running n individual backtests..

Introducing trading cost might decrease p
Air force
  • #365
  • Mar 11, 2026 4:53pm Mar 11, 2026 4:53pm
  •  amvt
  • Joined Feb 2025 | Status: Developer by Choice Trader by Heart | 202 Posts
Quoting W0lfram
{quote} {quote} I get it, but we don't have any params here; if IB_bull -> buy kind of trading logic. If it was ma cross strategy, then we should have to vary ma window running n individual backtests.. Introducing trading cost might decrease p
I get that, but you're probably misunderstanding what permutation tests do. I am not suggesting this to check for parameter overfitting, it was to validate whether the p-value itself is trustworthy or not.

The suspicious p-value of 0.0000 is a red flag because financial time series have properties that violate the assumptions of standard statistical tests like serial autocorrelation, volatility clustering and non-stationarity. A permutation test sidesteps all of these assumptions by asking "If I randomly shuffle the signal labels, how often do I get a result this good by chance?" It doesn't require i.i.d. returns, and it doesn't require you to have parameters to vary.

You are right about trading costs, but it's separate and would be a good idea anyways to check. But it does not replace permutation testing.
— Developed by MMQ
  • #366
  • Mar 11, 2026 4:56pm Mar 11, 2026 4:56pm
  •  amvt
  • Joined Feb 2025 | Status: Developer by Choice Trader by Heart | 202 Posts
Also, you might have a counter-argument like "but shuffling the labels destroys the time series structure." This is a common objection, and the answer is that you'd use block permutation (shuffling chunks rather than individual observations) to preserve local autocorrelation. Timothy Master's book "Permutation and Randomization Tests for Trading System Development" goes in detail about this.

This was pretty fun to discuss actually. Never had such intellectual discussions on FF, except with medici who I randomly came across on a forum and learned that he is a literal scientist lol
— Developed by MMQ
  • #367
  • Mar 11, 2026 5:07pm Mar 11, 2026 5:07pm
Quoting amvt
{quote} I get that, but you're probably misunderstanding what permutation tests do. I am not suggesting this to check for parameter overfitting, it was to validate whether the p-value itself is trustworthy or not. The suspicious p-value of 0.0000 is a red flag because financial time series have properties that violate the assumptions of standard statistical tests like serial autocorrelation, volatility clustering and non-stationarity. A permutation test sidesteps all of these assumptions by asking "If I randomly shuffle the signal labels, how often...
Quote
If I randomly shuffle the signal labels, how often do I get a result this good by chance?
So you want to swap buy/sell randomly?
Air force
  • #368
  • Mar 11, 2026 5:15pm Mar 11, 2026 5:15pm
  •  amvt
  • Joined Feb 2025 | Status: Developer by Choice Trader by Heart | 202 Posts
Quoting W0lfram
{quote} {quote} So you want to swap buy/sell randomly?
No, the returns stay fixed. You shuffle when the signal fires, not the trade direction.
— Developed by MMQ
  • #369
  • Mar 11, 2026 5:24pm Mar 11, 2026 5:24pm
Quoting amvt
{quote} No, the returns stay fixed. You shuffle when the signal fires, not the trade direction.
This sentence doesn't make any sense..
What do you shuffle (a deck of cards?) when signal fires if not the direction?
Order of trades? That's bootstrapping (what I was talking about)

p.s. Have your own examples?
Air force
  • #370
  • Mar 11, 2026 5:32pm Mar 11, 2026 5:32pm
  •  amvt
  • Joined Feb 2025 | Status: Developer by Choice Trader by Heart | 202 Posts
Quoting W0lfram
{quote} This sentence doesn't make any sense.. What do you shuffle (a deck of cards?) when signal fires if not the direction?
It takes 2 minutes to look up permutation test as described by timothy but a dumbed down version would be this.

You have two arrays, signal (1s and 0s) and returns. Normally you look at returns where signal=1. In the permutation test, you randomly shuffle the signal array and remeasure performance. Do this thousands of times. This tells you how often a random signal achieves your result. If your real signal beats 99.9% of random shuffles, it's probably real. If you still don't understand google's your best friend

Huh? How are you conflating bootstrapping with permutation?
— Developed by MMQ
  • #371
  • Mar 11, 2026 6:05pm Mar 11, 2026 6:05pm
Quoting amvt
{quote} It takes 2 minutes to look up permutation test as described by timothy but a dumbed down version would be this. You have two arrays, signal (1s and 0s) and returns. Normally you look at returns where signal=1. In the permutation test, you randomly shuffle the signal array and remeasure performance. Do this thousands of times. This tells you how often a random signal achieves your result. If your real signal beats 99.9% of random shuffles, it's probably real. If you still don't understand google's your best friend Huh? How are you conflating...
Quote
In the permutation test, you randomly shuffle the signal array and remeasure performance.
So you are swapping the alignment between signals.. Then one need to measure MFE and MAE for original signals. Look post#29
Who the fuck decided that term permutation is good for this? More like shift or offset or smth

Quote
You have two arrays, signal (1s and 0s) and returns.
You still think operation with 2 lists takes a lot of time?

Quote
How are you conflating bootstrapping with permutation?
If your question is "how sure am I about blah blah something" then you need bootstrapping

There is a whole bunch of stat tests and other econometric mental jerking that don't have practical significance tbh
Air force
  • #372
  • Edited 8:19pm Mar 11, 2026 6:48pm | Edited 8:19pm
Quoting amvt
{quote} It takes 2 minutes to look up permutation test as described by timothy but a dumbed down version would be this. You have two arrays, signal (1s and 0s) and returns. Normally you look at returns where signal=1. In the permutation test, you randomly shuffle the signal array and remeasure performance. Do this thousands of times. This tells you how often a random signal achieves your result. If your real signal beats 99.9% of random shuffles, it's probably real. If you still don't understand google's your best friend Huh? How are you conflating...
I'll take a look at Masters' book later.. maybe
Have you tried suggested approaches on practice with your backtests? Interesting how does this test handle 50 or 70%+ winrates.. Could you, theoretically, process the log if @eadwerdcap provided it or we can do it with synthetic data? I assume you have code templates already
Air force
  • #373
  • Mar 11, 2026 8:55pm Mar 11, 2026 8:55pm
  •  amvt
  • Joined Feb 2025 | Status: Developer by Choice Trader by Heart | 202 Posts
Quoting W0lfram
{quote} I'll take a look at Masters' book later.. maybe Have you tried suggested approaches on practice with your backtests? Interesting how does this test handle 50 or 70%+ winrates.. Could you, theoretically, process the log if @eadwerdcap provided it or we can do it with synthetic data? I assume you have code templates already
Would a good idea for you to do this since you obviously got more experience than me as you said it yourself x3.5. Since you didn't even know about permutation tests and confused it with bootstrapping, reading timothy would be a great idea aswell. Good Luck!
— Developed by MMQ
  • #374
  • Mar 12, 2026 12:35am Mar 12, 2026 12:35am
Excellent point about the first hour dynamics. I've found that the initial balance period is crucial not just for direction, but for risk calibration.

What many traders miss is adjusting position size based on the first hour's volatility expansion. If we see range compression in that initial period, I often reduce position size by 30-40% since breakouts tend to be more violent when they finally come.

The psychological aspect is huge too - those early moves create anchoring bias. Traders see a strong move in the first hour and assume the trend will continue all session, leading to oversized positions later in the day.

I track the ratio between first-hour range and average daily range across different pairs. EUR/USD typically shows 15-20% of daily range in first hour, while GBP/JPY can hit 35-40%. This helps me set realistic profit targets and stop distances.
FXOptimize.com — Free EA Portfolio Optimizer for MT4/MT5
  • #375
  • Mar 12, 2026 2:48am Mar 12, 2026 2:48am
I have to admit, the majority of the last 2 pages is above my pay grade
Can we state in layman's terms how all this talk of bootstrapping, permutations and synthetic data is of benefit to this simple IB breakout system?
What USEFUL statistical data would it provide that we can't get from just backtesting a m1 CSV with variables? (ok, I know testing with tick would be better).
I'm not criticising the skills of you two, but I'm sure I'm not the only one kind of lost with where this thread is heading.
  • #376
  • Mar 12, 2026 2:51am Mar 12, 2026 2:51am
Quoting amvt
{quote} Would a good idea for you to do this since you obviously got more experience than me as you said it yourself x3.5. Since you didn't even know about permutation tests and confused it with bootstrapping, reading timothy would be a great idea aswell. Good Luck!
Yeah right gpt guy
So you came here blabbering about technique you have haven't tried practically yourself but expect people to know what the heck you're talking about? I see..
Air force
  • #377
  • Edited 4:10am Mar 12, 2026 3:21am | Edited 4:10am
Quoting Danger-Mouse
I have to admit, the majority of the last 2 pages is above my pay grade Can we state in layman's terms how all this talk of bootstrapping, permutations and synthetic data is of benefit to this simple IB breakout system? What USEFUL statistical data would it provide that we can't get from just backtesting a m1 CSV with variables? (ok, I know testing with tick would be better). I'm not criticising the skills of you two, but I'm sure I'm not the only one kind of lost with where this thread is heading.
I fully agree, DM.
Common ways to do backtesting is ok and sufficient
Quote
There is a whole bunch of stat tests and other econometric mental jerking that don't have practical significance tbh

eadweardcap posted a detailed granularity stats, that's pretty it what one needs
I usually track floating PnL and DD (MFE/MAE) during each signal lifetime, if they're ~equal (usually) -> no point in further random tests, better to develop some exit strategy rules


Bootstrap simulations are good e.g when you can't take every single trade live, so you randomly sampling from historical PnL dataset
Air force
  • #378
  • Mar 12, 2026 4:55am Mar 12, 2026 4:55am
Quoting Danger-Mouse
I have to admit, the majority of the last 2 pages is above my pay grade Can we state in layman's terms how all this talk of bootstrapping, permutations and synthetic data is of benefit to this simple IB breakout system? What USEFUL statistical data would it provide that we can't get from just backtesting a m1 CSV with variables? (ok, I know testing with tick would be better). I'm not criticising the skills of you two, but I'm sure I'm not the only one kind of lost with where this thread is heading.
Same here. And that's not to invalidate W0lfram's work or anything. I just kinda lost it track of it. So this is the post that W0lfram is referring to:

https://www.forexfactory.com/thread/...1#post15603811

My train of thought was, well what if we can go back and see what's the most common thing that happens? When the most common thing that happens, what happens after that? What are the probabilities of those two things happening? Because in the end this is also a probabilities game right? So I just wanted to know what is the most probabilistic thing that happens and go off of that.

Everything beyond that, like the back testing, the montecarlo simulation and everything else I did with Python I feel was just additional stuff.
  • #379
  • Mar 12, 2026 5:04am Mar 12, 2026 5:04am
Quoting eadweardcap
{quote} Same here. And that's not to invalidate W0lfram's work or anything. I just kinda lost it track of it. So this is the post that W0lfram is referring to: https://www.forexfactory.com/thread/...1#post15603811 My train of thought was, well what if we can go back and see what's the most common thing that happens? When the most common thing that happens, what happens after that? What are the probabilities of those two things happening? Because in the end this is also a probabilities game right? So I just wanted to know what...
Only thing I would say about that post is that you did it with m5 candles (and I'm happy for Wolfie to step in and correct me if I'm wrong) but with a ORB system price can move very fast and far both ways within 5 mins..did it hit the target or the stop first? For pure price path we'd need tick but I think m1 should be accurate enough for this?
  • #380
  • Mar 12, 2026 5:11am Mar 12, 2026 5:11am
This is a back test of the following:

#USSPX500: 5 minute candles, from November 2022 to March 11th 2026 (yesterday). Why from november? 'Cause it's the most reliable data I can get in MT5.
Conditions:
Break Out Range: 16:30 - 16:44 (15min OR)
IF
> OR size is above 15 points and below 25 points
AND
> close of previous candle (1) is above the high of the OR high
THEN
> BUY Order.
> SL set at the OR low
> TP set at 1x of SL
> At 20:00 close the position (doesnt matter if it's winning or loosing)

Results:
Profitable Trades: 69 (58.97%) / Loosing Trades 48 (41.03%)
Drawdown (Max, Relative) 1.22%
Profit Factor: 1.28
Recovery Factor 1.65
Sharpe Ratio: 6.04
Attached Image (click to enlarge)
Click to Enlarge Name: TesterGraphReport2026.03.12.png Size: 92 KB
1 trader viewing now