null | Page 4 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Modifying indicator to account for NULL values - A Little Help, please! 8 replies

Attachments: null
Exit Attachments

null

  • #61
  • Feb 25, 2018 4:37pm Feb 25, 2018 4:37pm
Quoting hanover
Great thread, Sis. I wish I had your knowledge of statistics.
Thanks for the comment Hanover. With your coding experience and general market knowledge, I know deep down you already do somewhere in there
  • #62
  • Feb 25, 2018 4:44pm Feb 25, 2018 4:44pm
Quoting Sis.yphus
I know you know the assumptions you made were incorrect. Still wanted to share some thoughts on them. {quote} This won't ever be true since volatility is never constant. It is mean reverting, and is always fluxing. However, because it is mean reverting, you can confidently expect it to contract when it's been expanding for some time, and vice versa. So if your "windows" don't include all of the sessions, you will notice a big difference in the distributions during low and high vol sessions. {quote} It' low for a bullish candle almost always forms...
Thanks for clarifying these points. I'm working on it and will post my results for tick data tomorrow. Maybe then we can discuss how to use this data. I've thought that the best way would be looking at the 5-dimensional probability distributions (probability of observed pit/peak time and prize distance as well as #of 0-line crosses and 0-line times).
  • #63
  • Feb 25, 2018 6:36pm Feb 25, 2018 6:36pm
Quoting AntiCre
{quote} Thanks for clarifying these points. I'm working on it and will post my results for tick data tomorrow. Maybe then we can discuss how to use this data. I've thought that the best way would be looking at the 5-dimensional probability distributions (probability of observed pit/peak time and prize distance as well as #of 0-line crosses and 0-line times).
Looking forward to it. Don't limit yourself to tick data either imo.

Heres what I usually see in terms of the number of 0 line crosses.
Attached Image
  • #64
  • Edited 5:36pm Feb 26, 2018 5:26am | Edited 5:36pm
Quoting Sis.yphus
Don't limit yourself to tick data either imo.
How shall I interpret it? Do you refer to the trend definition on M1 (or even higher) candlesticks?
BTW - I don't think that you use the multidimensional distribution but are more after pattern recognition/classification (something like: if prize crossed 0 between [x1,x2] ticks and pit distance within [y1,y2] and open line has been already crossed between [z1,z2] times and trade is in trend direction, [t1 > 0.8], then trade).

Now I changed my code to correct the wrong assumptions. This time I looked at GU ticks for the time between 09:00 and 10:00 (GMT) over two weeks. The sample size is > 50,000 and thus imho big enough to obtain reliable measures.

Here is what I exactly did:

  1. Start with i-th tick and note it's prize (open)
  2. Look at the (i+499)-th tick and not the prize (close)
  3. Determine the highest and lowest tick prize for the interval [i:i+499]
  4. Determine if the high or low occured first
  5. Calculate the open wick length and the close wick length (open < close: open wick length = open - low, close wick length = high - close ; and vice versa for open > close)
  6. Note the time for the high and low (measured in the #ticks occured since bar i)
  7. Count how often the prize crosses the open prize (in the direction of the high resp. the low, depending on which occured last)
  8. Determine the time (measured as tick time again) of the last open cross in the direction of the occurence of the last peak/bottom (high/low)
  9. Repeat the procedure for the next tick (the tick i+1]

The distributions of the 6 measures is shown below. Since I'm interested in trading open line crosses I deleted the 0 time/prize occurences from the stats and only printed the histograms for times, #occurences, and prizes > 0.

Although the results look similar to the images Sis.yphus posted on page 1, I can see some differences. Additionally there are some differences that I still don't understand. Why show the peak and pit distance distributions the same x axis as the other plots? Why is the peak time/distance distribution very different in comparison to my Close wick distributions? What about the difference in 0 line crosses and open line crosses?

If anyone is interested in my R code I can post it here ...

Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot1.png Size: 18 KB
  • #65
  • Feb 26, 2018 1:18pm Feb 26, 2018 1:18pm
Inserted Video


Really interesting interview.
  • #66
  • Edited Feb 27, 2018 7:37am Feb 26, 2018 2:27pm | Edited Feb 27, 2018 7:37am
The most simple application for the open/close wick time distribution is imho a quite old idea. I and others interpreted it in one of the oldest running gag @ FF (AMAZING strateg by Merlin).

Since the open wick (which is per definition a retracement in the counter direction of a candlestick) occurs with high probability in the first half and the close wick in the second half of a candle's lifetime, you could just look at the high/low created during the first half and then buy/sell if that high/low is exceeded. However, that requires that the occurences of open/close wick times are independent from each other, so that the occurence time distribution is really in our favor. Thus, I created a 3D histogram to show the relation of open and close wick times. It is in our favor, thus the idea could be applied (although this is not the intention I have with the analyses). However, it is a logical explanation why the separation of a day/hour/whatever TF into two subdivisions could be useful ...

PS: ... and if none of the last x/2 candles went above/below the highest/lowest line, then it's often a good idea to trade in the direction of the lowest/highest line.
Attached Image(s) (click to enlarge)
Click to Enlarge Name: Screenshot1.png Size: 22 KB
Click to Enlarge Name: Screenshot2.png Size: 16 KB
  • #67
  • Mar 1, 2018 9:24am Mar 1, 2018 9:24am
Strengthen Trading Psychology with Expected Value - How to Never Lose Again

This is kind of a continuation of the revenue requirements post. I briefly touched upon expected value there in terms of reaching your daily goals but today I wanted to talk about the benefits of knowing your expected value has on your trading psychology, especially when it comes to trading low hit systems.

For those who forgot:

Expected Value = (Reward * Win Rate) - (Risk * Lose Rate)

Now in order for this number to be accurate, you need to really verify your system so you can trust the data it gives you.

This number is beautiful. When accurate, it's your psyches best friend, however, if inaccurate, it can be your worst enemy.

This number essentially tells you what you make each time you place a trade with that system, but like an LTV(Life Time Value of a Customer) for a service business or something similar. The LTV doesn't mean you will make that much off that exact customer, but it does mean, that is what you make for each new customer brought in over time.

Same with EV. The EV has nothing to the outcome of any single trade. It simply represents what you make for each trade placed over time.

One system I trade has an expected value of 42% whatever the risk on a trade is.

This provides huge conviction when trading because I know as long as I enter the set up correctly, regardless of the outcome of that particular trade, I am adding 42% of whatever the risk is back into my account, at some later point in time.

If you know your numbers and they are accurate, you never actually have a losing trade.
  • #68
  • Mar 1, 2018 9:25am Mar 1, 2018 9:25am
Looks like some huge buy orders are hitting EURJPY. GU GJ EU all sliding while EJ plows up...

Interesting.
  • #69
  • Mar 2, 2018 9:08am Mar 2, 2018 9:08am
GA has been pretty fun this morning. Took it short last night and now riding it long until the weekend.

Also, long some UC to see what happens and looking to buy a dip in AU and AN.

Something I have been thinking about again (and this comes up often) is whether or not to hold trades through the weekend.

At times I feel no need to as it's a whole other level of risk(gaps past your stops = major pain). On the other hand, though, a lot of my models don't consider "weekends" and treat it all the same so I often wonder if I am leaving pips on the table by avoiding opportunities I would have normally held through the weekend.
  • #70
  • Mar 2, 2018 9:34am Mar 2, 2018 9:34am
Good Trading is Boring

Despite what the movies and documentaries show you, good trading is BORING! Not like how it used to be in the Open Outcry Markets. Then, trading was exciting, a lot of yelling, quick action, and reading the crowd.

Nowadays as we all sit behind computers, the game has changed (I hope you are not yelling at your computer). If your trading is exhilarating, or your heart races everytime you get in the market, you are doing it wrong.

Your trading should not be "exciting" or get your heart racing. It needs to be business as usual. Unfortunately, this is really hard to do if you don't understand your system inside and out.

If your trading your system isn't "boring" to trade on some level, then I'd argue there are parts of your system that need work or you just like gambling

Nowadays for me, essentially none of the actual "trading" is handled by me. I enter with limit orders and have an EA manage the open and pending orders for me. All I do is find where and when to place orders and place them... Business as usual...
  • #71
  • Mar 2, 2018 9:48am Mar 2, 2018 9:48am
Quoting Sis.yphus
Good Trading is Boring Despite what the movies and documentaries show you, good trading is BORING! Not like how it used to be in the Open Outcry Markets. Then, trading was exciting, a lot of yelling, quick action, and reading the crowd. Nowadays as we all sit behind computers, the game has changed (I hope you are not yelling at your computer). If your trading is exhilarating, or your heart races everytime you get in the market, you are doing it wrong. Your trading should not be "exciting" or get your heart racing. It needs to be business as usual....
Sis.yphus,

How often do you check the status of trades, i.e. how many executed, how much profit or drawdown etc...?

If you don't mind me asking, How do you handle rest of the time when you are not hands on trading?

Thanks
  • #72
  • Mar 2, 2018 9:55am Mar 2, 2018 9:55am
Quoting aaven
{quote} Sis.yphus, How often do you check the status of trades, i.e. how many executed, how much profit or drawdown etc...? If you don't mind me asking, How do you handle rest of the time when you are not hands on trading? Thanks
I look at the market and watch it almost all the time so I am monitoring my trades ALOT. Some could probably argue too much.

To pass the time, I am almost always working and testing new ideas every day, I like to discuss ideas with trading buddies on discord, share some thoughts on FF, and usually by the afternoons when things slow down, I check out till later that night to spend time with the family, workout, etc.
  • #73
  • Mar 2, 2018 10:06am Mar 2, 2018 10:06am
Quoting Sis.yphus
{quote} I look at the market and watch it almost all the time so I am monitoring my trades ALOT. Some could probably argue too much. To pass the time, I am almost always working and testing new ideas every day, I like to discuss ideas with trading buddies on discord, share some thoughts on FF, and usually by the afternoons when things slow down, I check out till later that night to spend time with the family, workout, etc.
Thanks Sis.yphus...
  • #74
  • Mar 2, 2018 12:39pm Mar 2, 2018 12:39pm
I worked very hard during these days. Here is how far I came (today's GU M1) ...

The teal dots represent buy and the red represent selling opportunities based on my stats. (Exit logic not yet implemented.) But there's still one problem - the non-linear trend is quite lagging. Thus, I implemented another trend detection tool. Now the lag is (a bit) less but still observable ... The best opportunities are imho the first appearences of a teal or orange dot.

Any recommendations to improve it?
Attached Image (click to enlarge)
Click to Enlarge Name: Screenshot1.png Size: 35 KB
  • #75
  • Mar 2, 2018 3:59pm Mar 2, 2018 3:59pm
Quoting AntiCre
I worked very hard during these days. Here is how far I came (today's GU M1) ... The teal dots represent buy and the red represent selling opportunities based on my stats. (Exit logic not yet implemented.) But there's still one problem - the non-linear trend is quite lagging. Thus, I implemented another trend detection tool. Now the lag is (a bit) less but still observable ... The best opportunities are imho the first appearences of a teal or orange dot. Any recommendations to improve it? {image}
Looks really interesting.

Can't tell exactly what is going on based on the image but here is an idea I had.

I would look at the accuracy of each dot and see how time decay with a continued trend affects it. If the prob of the first dot is correct is the same as the second, third, so on.

This way you can find a cut off point of when the signals are no longer valid or at statistically less likely to be valid.

I have a hunch on how the accuracy will decay over time.

Also, I think you will need some strict exit rules so you have a target. Getting a good entry is key, but it's also just as important to have good exits.
  • #76
  • Mar 5, 2018 9:21am Mar 5, 2018 9:21am
Drop Offs & Balance Points

We have a problem... a birthday problem.

Inserted Video


As you move through each day of the year, the number of options remaining for that person's birthday decreases, and the probabilities adjust accordingly.

Let's imagine a ball being kicked in 1 direction 1000 times.

50% of the time it goes 50 feet.
75% of the time it goes 100 feet.
87.5% of the time it goes 200 feet
93.5% of the time it goes 400 feet.

Now imagine we are trying to estimate how far the next ball will travel in real time as it makes its move.

Before the kick of the ball the odds are:
50/50 we go 50 feet
25/75 we go 75 feet
12.5/87.5 we go 100 feet
6.5/93.5 we go 150 feet.

However, once the ball is traveling the odds become very different.

So price makes it 50 feet.

What are the odds price will go 75 feet? Intuition says 25%, but it's really 50/50 again as we are at a new "balance point".

So price makes it 75 feet.

What are the odds price will go 100 feet? 12.5%? Nope, it's again 50/50 once the ball has already made it 75 or more. We are at another "balance point".

Simple enough right?
  • #77
  • Edited 9:51am Mar 5, 2018 9:35am | Edited 9:51am
[quote=Sis.yphus;10840533]Drop Offs & Balance Points We have a problem... a birthday problem.
Inserted Video
As you move through each day of the year, the number of options remaining for that person


interesting, the key figure is 50 especially if it is points gained in a forex movement. We are speaking Frequency Distribution...

edit... as you used 50 (kicks) i saw candle tails 5+ the occasional 0(loss) mand + 10, + 20. + ? never average them as you will lose, but the 5 is the key...

To get even better results, if the stat is 5 , trade for 2 or 3 or 4 , this way you can hit the trade harder and even COMPOUND yaaaaaaay, my baby

edit, you can compound anyway
SEE LINE,TRADE LINE..PRICE HAS TO GO SOMEWHERE,,, PRICE WILL GO SOMEWHERE.
  • #78
  • Mar 5, 2018 9:36am Mar 5, 2018 9:36am
[quote=billytt;10840625]
Quoting Sis.yphus
We are speaking Frequency Distribution...
Seems to be the only language I want to speak these days...
  • #79
  • Mar 5, 2018 9:50am Mar 5, 2018 9:50am
[quote=Sis.yphus;10840632]
Quoting billytt
{quote} Seems to be the only language I want to speak these days...
i was using these stats for ages and did not know they had a name, but a guy who visited my thread said quote,, that is FD, you learn something new every day...
SEE LINE,TRADE LINE..PRICE HAS TO GO SOMEWHERE,,, PRICE WILL GO SOMEWHERE.
  • #80
  • Mar 5, 2018 11:47am Mar 5, 2018 11:47am
Quoting Sis.yphus
Drop Offs & Balance Points We have a problem... a birthday problem. https://www.youtube.com/watch?v=KtT_cgMzHx8 As you move through each day of the year, the number of options remaining for that person's birthday decreases, and the probabilities adjust accordingly. Let's imagine a ball being kicked in 1 direction 1000 times. 50% of the time it goes 50 feet. 75% of the time it goes 100 feet. 87.5% of the time it goes 200 feet 93.5% of the time it goes 400 feet. Now imagine we are trying to estimate how far the next ball will travel in...
Great explanation. That's exactly the problem with the 5-linear subsequence problem from EURUSDD. Although you will see at least 1 of the 4 conditions fulfilled 93.75 % of the time, the probability of P(fulfilled with 5th nmber|not yet fulfilled with the 4th number) is still 0.52 ...
1 trader viewing now