Questions for real pros: How to trade with reference points only? | Page 4 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Price action trading - Trade like the pros! 964 replies

Retail Trade Gains vs Pros 6 replies

newbie questions on daily pivot points 18 replies

Retail Sales? Trade Balance? What it means? Newbies ask here. Pros answer here. 9 replies

Attachments: Questions for real pros: How to trade with reference points only?
Exit Attachments

Questions for real pros: How to trade with reference points only?

  • #61
  • Jul 6, 2023 11:42am Jul 6, 2023 11:42am
Quoting AntiCre
{quote} Allow me to complement this. Under the assumption of a random walk market I wanted to know which kind of candlesticks I can expect on average. Thus, I considered a random walk with (-1, +1) increments and generated 10000 candlesticks with 10000 steps. Here are the results: {image} On the x axis you can see the probability for getting a +1 in any of the 10k step. The top panel shows the average bull bars that can be expected for every probability, the middle panel the average bear bars that have shown up in my simulation and in the bottom...

I think the pressure is always essentially “equal”.

with this simulation, I am assuming the outcomes produce a normal distribution with 3 kurtosis?

can you run the simulation, and with a kurtosis of 4-5 to see how that changes the candles? I imagine it will make them more unstable, with much more overlap.
ri · skuht
  • #62
  • Jul 6, 2023 11:46am Jul 6, 2023 11:46am
Quoting Riskcuit
{quote} I think the pressure is always essentially “equal”. with this simulation, I am assuming the outcomes produce a normal distribution with 3 kurtosis and 0 skew? can you run the simulation, and with a kurtosis of 4-5 and 0 skew to see how that changes the candles? I imagine it will make them more unstable, with much more overlap.
Well, I could do it tomorrow. But remember, this is just a coin tossing simulation. Thus, not really a skew or curtosis ...
  • #63
  • Jul 6, 2023 1:38pm Jul 6, 2023 1:38pm
Quoting AntiCre
{quote} Well, I could do it tomorrow. But remember, this is just a coin tossing simulation. Thus, not really a skew or curtosis ...
will be interested to see the result.

I’d also look at plotting the outcome (final value)of each 10000 step candle as a histrogram. I imagine the kurtosis is 3 and the distribution is normal so the kurtosis is there.

Quoting apprendista
Nice to see the thread back on! Nice work you shared Anti. I wanted to ask you something, I'll do here since maybe others can benefit also. So you said: {quote} My question is: why do you consider the market like a random walk? There's similarities of course but from my researches market better resembles a Gaussian walk (Brownian motion), vast majority of industry also rely on that model. The differences between the two are many, but the most relevant to our topic is momentum/trend; in Gaussian walk the model (on the most basilar form, the arithmetical...
another note on this. Gaussian is a good example of something similar to price only on a tick by tick basis. But it fails to really capture how price behaves in the big picture.

A example to illustrate this.

Gaussian always has kurtosis of 3. Markets do not, their kurtosis remains above 3 regardless of time horizon. This where people get annihilated building models purely on the assumption of a Gaussian distribution.

Attached Image (click to enlarge)
Click to Enlarge Name: DD08C731-B582-4574-A539-E948539B4774.jpeg Size: 125 KB


people usually overlook how large the impact of this excess kurtosis can have on tail events. For example, a single daily drop of 228 pips (open to close), has roughly about a .01% chance of happening with AUD/USD’s mean and STD if you modeled a Gaussian distribution with it. In reality though, this occurs with roughly a 0.37% probability, meaning it happens 37x more often than it would under a Gaussian distribution.
ri · skuht
  • #64
  • Jul 7, 2023 3:12am Jul 7, 2023 3:12am
Quoting Riskcuit
{quote} I think the pressure is always essentially “equal”. with this simulation, I am assuming the outcomes produce a normal distribution with 3 kurtosis? can you run the simulation, and with a kurtosis of 4-5 to see how that changes the candles? I imagine it will make them more unstable, with much more overlap.
Shit. I already found a way to produce distributions for given means, variances, skewnesses and kurtosis. However, I cannot find the R code that I've used to produce the image posted yesterday anymore. Moreover, how should I have modelled the advantage for one side? Just moving the mean wouldn't be the pendant to an advantage for +1 in comparison to -1. Changing the skewness would have of course created different candles ... Sorry, since I seem to haven't saved the R code I think I won't do it ... I mean, you can already imagine the result for both alternatives discussed above without executing any code.
  • #65
  • Jul 7, 2023 8:27am Jul 7, 2023 8:27am
Quoting AntiCre
{quote} Shit. I already found a way to produce distributions for given means, variances, skewnesses and kurtosis. However, I cannot find the R code that I've used to produce the image posted yesterday anymore. Moreover, how should I have modelled the advantage for one side? Just moving the mean wouldn't be the pendant to an advantage for +1 in comparison to -1. Changing the skewness would have of course created different candles ... Sorry, since I seem to haven't saved the R code I think I won't do it ... I mean, you can already imagine the result...
No worries, If you had the code and just needed to tweak some inputs, would have cool but not worth re-rolling the script. Like you said, it’s pretty intuitive what the result would be, more over lap between candles around the mean with longer wicks/large “trend” candles.

was thinking about your original post, specifically this and wanted to share some thoughts.


“and how one can escape the trade-off between risk-reward (RR) and the probability of winning. For example, how can one expect a higher probability of winning than 0.5 with a 1:1 RR ratio?”

I have come to the conclusion that you can’t gain a consistent edge using fixed RRs if there is no skew in the data itself. There is temporary skew in each price path, but the aggregate of all of them results in essentially 0.0 so it goes back to need to filter (assuming it’s even possible to consistently do so).

if you want to only use fixed RRs, you need skew in either price itself as a whole or in the ratio between touching and closing above any price (value of 2 means no skew). Think about casino games, they offer many RRs on a roulette wheel cause of the skew, they don’t care what RR they take. This is the only way they can work mathematically.
ri · skuht
  • #66
  • Sep 19, 2023 3:40pm Sep 19, 2023 3:40pm
Quoting AntiCre
{quote} You're just allowed to start own threads after you've made 20 reviewed posts. However, please search a thread where your question fits better - not here. This thread was intended to investigate trading opportunities around reference points!
maybe it is better to see again the Pi thread by CP.
  • #67
  • Edited 6:04am Dec 3, 2023 5:53am | Edited 6:04am
Since I read about it the first time, I always thought that Fibonacci numbers in the markets are esoteric hocus pocus. However, today I did an experiment. I created candlesticks that contained arbitrarily 10000 data points (ticks). Whenever such a candlestick was bullish, I analyzed every difference with a lag of 2500 ticks inside these 10000 tick-candlesticks and recorded i) the maximal retracement (i.e. minimal distance between every price and the price 2500 ticks ago), ii) the maximal pullback (i.e. maximal distance between every price and the price 2500 ticks ago) as well as iii) the length of the upper wick inside these 10000 tick-candlestick.

What I found was this:

  1. The ratio of the average maximal pullback to the average maximal retracement approximates the golden ratio, phi = 1.618.

Just want to give some food for thought ...

BTW: I get very similar ratios for the above mentioned measures if I use random walks with (-1, 1) and normally distributed increments instead. To me this indicates again that the market indeed behave like a random walk (at least on average).

  • #68
  • Dec 3, 2023 5:01pm Dec 3, 2023 5:01pm
Quoting AntiCre
Since I read about it the first time, I always thought that Fibonacci numbers in the markets are esoteric hocus pocus. However, today I did an experiment. I created candlesticks that contained arbitrarily 10000 data points (ticks). Whenever such a candlestick was bullish, I analyzed every difference with a lag of 2500 ticks inside these 10000 tick-candlesticks and recorded i) the maximal retracement (i.e. minimal distance between every price and the price 2500 ticks ago), ii) the maximal pullback (i.e. maximal distance between every price and the...
Hey mate, love your posts and the quant analysis of candlesticks, I've also done some work down this path aswell, would love if you continue posting your findings, very interesting stuff. I also came to the conclusion that the market is essentially random, when testing random entry's I found that 1:1 rr produces almost bang on 50% winrate but when I started to influence my entry's the winrate drops. I also enjoyed cp's posts regardless if I found them valid also
  • #69
  • Edited 2:30am Jan 4, 2024 1:18am | Edited 2:30am
Dear friends,

happy new year to all of you!
I'd like to invite you to continue our discussion on candlestick's reference points.
Before I start, I'd like to repeat some of my current beliefs on the market:

  1. The market behaves (at least in the long term) like a random walk. Yes, there might be times when the market deviates to be random which may manifest itself as either being flat ranges or strong trends.
  2. The market behaves that random way, because it is a dynamical and chaotic system. These beliefs bring a multitude of assumptions and "painful" conclusions:

    1. The chaos in the price process makes it noisy.
    2. The long-term predictability of the market is derailed by ever-changing initial conditions and small (an bigger) influences on the market (e.g. initial price, average market participant's beliefs, interest rates, fundamental data, etc.)
    3. Thus, the price is not predictable!

Of course, this statement is arguable and highly depends on everyones own belief. But here I'd like to only show my own beliefs. Thus, I'll stick to this line of thoughts.
If the market (price development) is unpredictable in general, how could one profit from the market movements?
To me this was long time the trickiest question on earth. I don't lie at you if I tell you that it took me years to "discover" this myself. Only after intense research I was able to grasp this. And now I know that so many people here and in other places were right when the regurgitated: "Follow the market, don't predict it." or "Follow the trend".
Even though the price process is viewed as dynamic and chaotic, there are still some "repeatable" patterns: candlesticks. This is where my candlestick investigations started: I thought it must be somehow possible to exploit this cyclic or semi-cyclic price moves to later use it to build a system to follow the market.
Yes, I could have used M1 data or candlesticks from any other TF. However, if I'd like to get a comparable measure, every candle should at least contain the same amount of price updates (ticks; not in the sense of price differences!). A measure, that "re-adjust" with every new incoming tick. Thus, summarizing a number of ticks as a candlestick and the same number of candlesticks with a "candlestick frame" moved 1 tick to the right as another candlestick, etc.

Here's what I've done:

  1. Look at a "price frame" (definition: a number of price updates") of 1000 ticks.
  2. View these 1000 ticks as 1 candlestick. Any candlestick can form in different ways:

    1. directional bias

      1. bullish
      2. bearish
      3. doji

    2. strength of directional bias (use a "useful" classification of the strength of your candlestick; let the data guide you in your choice; why not looking a an average bullish/bearish candlestick?)

      1. weak
      2. medium ("normal")
      3. strong

    3. order of high/low

      1. time of high > time of low
      2. time of low > time of high (with "time" meaning number of ticks in your price frame's candlestick; best expressed in percent)

For every combination (bullish, weak, time of high > time of low; bullish, medium, time of low > time of high; ...; doji, time of high > time of low; ...; bearish, strong, time of low > time of high) calculate the average

  1. open - close distance
  2. upper wick
  3. lower wick
  4. time of high (expressed in percentages to your price frame used)
  5. time of low (expressed in percentages to your price frame used)

Repeat the process for at least two bigger or smaller price frames.

Questions:

  1. What does the average bullish and how does the average bearish candlestick look like?
  2. Which trends/tendencies do you see?

    1. What happens to the wicks/bodies/average high/low times from weak to strong trends?
    2. How often (expressed in percentages) do th different types of candlesticks you've classified appear?
    3. How do number of appearances of dojis and wick-to-body ratios change for different price frames?

I believe it's crucial to collect these stats yourself to really see the value in it. Which values can you find? - Either constants (which is great because it will tell you that you don't need to evaluate thresholds/cut-off points for different price frames you're looking at) or trends (which is also great because you may be able to exploit these tendencies to better understand the market's trends or ranges).
I already found answers to these questions and will only reveal what I found if you'll show your investigations (it doesn't need to be the same way I've done it but I'd like to see you were/are working on this ).

  • #70
  • Jan 4, 2024 3:38am Jan 4, 2024 3:38am
Quoting AntiCre
Dear friends, happy new year to all of you! I'd like to invite you to continue our discussion on candlestick's reference points. Before I start, I'd like to repeat some of my current beliefs on the market: The market behaves (at least in the long term) like a random walk. Yes, there might be times when the market deviates to be random which may manifest itself as either being flat ranges or strong trends. The market behaves that random way, because it is a dynamical and chaotic system. These beliefs bring a multitude of assumptions and "painful"...
there is a tick frame in ctrader. Is it good to use?
  • #71
  • Jan 4, 2024 8:04am Jan 4, 2024 8:04am
Quoting Karlorenz14
{quote} there is a tick frame in ctrader. Is it good to use?
I never used ctrader. But I think it should work, too. To not see the general beavior of prices anymore, your broker really needs to manipulate the data feed extensively.
  • #72
  • Jan 4, 2024 8:13am Jan 4, 2024 8:13am
Quoting AntiCre
{quote} I never used ctrader. But I think it should work, too. To not see the general beavior of prices anymore, your broker really needs to manipulate the data feed extensively.
are you profitable on this strategy? Do you use tools or only manually?

what i actually doing right now is used the 1st few hours of session opening as a reference points to see the intentions and directions. I would try to study the one you mentioned above.
  • #73
  • Jan 4, 2024 11:05am Jan 4, 2024 11:05am
I traded a demo account with multiple strategies derived from these statistics. Hey, if your strategy works, then kudos to you! Then I think there's no need to jump on something else (of course, you're free and more than welcome to do the work).
  • #74
  • Jan 5, 2024 3:38am Jan 5, 2024 3:38am
Quoting AntiCre
I traded a demo account with multiple strategies derived from these statistics. Hey, if your strategy works, then kudos to you! Then I think there's no need to jump on something else (of course, you're free and more than welcome to do the work).
if it has winning rate and high RR, then im more interested
  • #75
  • Jan 5, 2024 12:22pm Jan 5, 2024 12:22pm
Quoting Karlorenz14
{quote} if it has winning rate and high RR, then im more interested
Dear Karlorenz14,

this is not just directed to you. But maybe I should state it bold in here: I have put this thread into the category "Trading Discussion" because I was/am mostly interested in understanding how trading can work. I'd like to know what are the main aspects that allow a trader to get more money from the market than giving to it (or her broker).

Thus, I'm not going to discuss how I applied some theories. Well yes, I'm going to do this, but not exhaustively. I just like to discuss some core concepts and present some ideas to those who are willed to learn and do a bit of extra work. This is not a thread with a super simple indicator which will make you rich soon! This thread is also not intended for discussions about trading ideas, closed/opened trades and equity curves. It will be just theoretical gibberish

But if you stay encouraged and attend focussed in the discussion, you hopefully may see some bits and pieces that could help you to build your own strategy.
  • #76
  • Jan 6, 2024 5:02am Jan 6, 2024 5:02am
Hi AntiCre!

Quoting AntiCre
Throughout the years, I've consistently observed that the way prices move is not significantly different from random walks. For example, the expected profit is strongly correlated with the holding time/trading duration.
Price moves because of market orders executions. Market is not random. It moves because of money. Money makes money. If there will be no market orders there will be no movement and limit orders in the orderbook will stay there forever. Also stops are market orders too (always), so triggering stops also cause market order executions. This is the basics. There is no sense to dream or create ideas how it works. It works this way only and always.

Quoting AntiCre
Therefore, I want to go back to the basics of trading and solely focus on reference points (such as prices of an asset from 0, X, Y, and Z minutes ago or the OHLC of a higher rolling candle) in my learning process.
That's why this statements (or idea) make no sense at all. There are no reference points. The past will never show you the future. The past is the past. It was. The only one thing you can use the data from the past is to calculate the amount of stops to figure out where is cheaper to go. Candles itself are very dangerous fake. Especially open and close values. Because this is the average for certain period. Open price and close price is fake data 100%. Since the market moves with different speed (volume) this is nonsense to create stable x-min candles and use open/close values from them. It just average the data to total mess. All you can use is high and low. Also remember about sessions, gaps, weekends and so on. All add move mess to ohlc and the data it totally corrupted at the end.

Im my opinion (based on years of algo trading) the only thing you can rely to in trading is volume. There are no support and resistance levels, there are no ohlc data, there are no titeframes. There are no candles, patterns and other magic things everyone can see just using eyes. Only Volume. Real trades. Orderflow. Even the orderbook is fake and you can't use it to make trading decisions (I can tell you move how they scam the orderbook, I'm doing this too btw it called spoofing). So only volume is the thing.

Stay greeen!
Observer effect
  • #77
  • Jan 6, 2024 5:20am Jan 6, 2024 5:20am
Quoting ryuryu
Price moves because of market orders executions. Market is not random. It moves because of money. Money makes money. If there will be no market orders there will be no movement and limit orders in the orderbook will stay there forever. Also stops are market orders too (always), so triggering stops also cause market order executions. This is the basics. There is no sense to dream or create ideas how it works. It works this way only and always. {quote} That's why this statements (or idea) make no sense at all. There are no reference points.
Well, I didn't want to open the question of randomness or determinism (it has been discussed dozens of times on this forum). I can see in your explanation that the market can be viewed as deterministic. Yes, traders build orders which cause price to move. However, imo the size of the traders submitting their orders, the order sizes and the times of orders are indeterministic, and thus random to me. To me this is the explanation why the average candlestick shows almost the exact same wick-building times and wick-to-range/body-to-range ratios than random walks with arbitrarily chosen symmetric increments. To me that's an indirect prove for market being random. But yes, there might be times when the market is deterministic or even predictable to some degree. Unfortunately not for me

Quoting ryuryu
The past will never show you the future. The past is the past. It was. The only one thing you can use the data from the past is to calculate the amount of stops to figure out where is cheaper to go.
Absolutely agree!

Quoting ryuryu
Candles itself are very dangerous fake. Especially open and close values. Because this is the average for certain period. Open price and close price is fake data 100%. Since the market moves with different speed (volume) this is nonsense to create stable x-min candles and use open/close values from them. It just average the data to total mess. All you can use is high and low.
Maybe you're right on this (never thought about it) - but only if you use prices as fixed reference points. What I'm looking at are OHLC of candles rolling on a tick-by-tick base. But even if I wouldn't use rolling candlesticks, the candles' reference points (OHLC) give some measures on what can be expected during a particular period of time, i. e. how much movement can be expected? How much counter-movement during the same time period can be expected? Etc.

Thanks for passing by and your warnings! I think such a word is important. I (and probably most beginners) did the same mistake in the beginning. However, I hope you can see right now that I don't use open and close as fixed reference points, but as part of measures to see what can be expected (or what has happened during the last x ticks/minutes/hours).
  • #78
  • Jan 6, 2024 6:24am Jan 6, 2024 6:24am
This is very big discussion. Let me explain with just one example your "candle symmetric".

First we have to check what candle is. It has open, high, low and close values.
Now what is open? It is the nearest price from the start of the selected period. Let's say it is 12:00:00 and we have 1 hour timeframe. So open can be the trade from 12:00:01 to 12:59:59. Pretty big gap, isn't it?
What will use see in both cases? Same colored rectangle called candle. And the candle size will be equal in both cases. If there were just 2 trades or 2000 trades inside you will see exactly the same candle.
Next you add your favorite moving average indicator, for example. In fact it doesn't matter how it called. They all are using ohlc data. So this is just for example. So you are trying to make average using candles with 2 trades and with 2000 trades. What will be the average? It depends where is you now. If at the end and both are +- equal it is ok. If at the start when 2000 trades candles begins then... then you will have corrupted data. Same with gaps. Do you shift your data after gaps? Nope. So how your magic moving average can give you correct results? It is just not possible.
I'm now working hard with LLM and big data. And I can say that data, correct data, is very important. No, not very, it must be correct. If it is not good then you will fail.
Using candles in trading, especially from higher timeframes is very dangerous. Since you average the corrupted data more and more and at the end you will get total useless mess.
People say that candles was used in Japan to count rice. Yes. But how you think will it be possible to sell x100 rice than last year? x1000? Nope. Is there were stops? Nope. Leverage? Heh yeah, x100. But now it is possible to increase the volume 100 and 1000 times.

But you correct. Open and close are fixed reference points. The only problem that you can't reference to it. Better use high and low. At least they are high and low for selected period. But open and close are totally synthetic.
Observer effect
  • #79
  • Jan 6, 2024 6:49am Jan 6, 2024 6:49am
Attached Image (click to enlarge)
Click to Enlarge Name: 123.png Size: 30 KB


found this could be helpful. i make no comment about others on these forums anymore. not worth it. equations in screenshot have worked consistently since rice trading used candlesticks. a person creates their own trading wisdom. its always about why you do it.

have fun.

O.
protein for breakfast !
  • #80
  • Jan 6, 2024 6:52am Jan 6, 2024 6:52am
Quoting ryuryu
This is very big discussion. Let me explain with just one example your "candle symmetric". First we have to check what candle is. It has open, high, low and close values. Now what is open? It is the nearest price from the start of the selected period. Let's say it is 12:00:00 and we have 1 hour timeframe. So open can be the trade from 12:00:01 to 12:59:59. Pretty big gap, isn't it? What will use see in both cases? Same colored rectangle called candle. And the candle size will be equal in both cases. If there were just 2 trades or 2000 trades inside...
I agree again. You don't know how many "price changes" you had during the lifetime of a candlestick at just looking at price (well, you could use tick volume). However, to be consistent, I suggested in #post 69 to use 1000 price changes (ticks in the meaning of price updates, not 1/10th of 1 pip). That way one can exclude inconsistent measures. And in this case, if I'd apply your MA, I'd always get an "equal" estimate. But even using candlesticks may give an idea on how much price moves on the average tick number in the FX markets since the main idea was to come up with an average candlestick by applying means to different measures.

Tbh, I haven't mentioned weekend gaps here. But I assumed that those who are working with data know to exclude it. Well yes, you can have other gaps in your data (even assuming you're having a perfect data feed). But these gaps have a more or less constant distribution and are symmetrically distributed around 0. Thus, if the calculation period of my rolling 1000-tick candlesticks is big enough, I should (of course only on average) see an (almost) equal number of up and down gaps, i.e. price changes from one price update to another).

That way we've eliminated the 1st issue (different tick numbers in candlesticks; using candlesticks defined by price updates, not time cutoff points) and accounted for the 2nd issue (dealing with gaps). Haven't we?
0 traders viewing now