I wonder if there has ever been a pattern of 10 green candles followed by 5 red candles before. This "10 green, 5 red" pattern seems relevant now: we just observed 10 green candles between 21-Aug to 30-Aug-2021 and we are now seeing 3 green candles in a row 31-July to 02-Aug. It would be interesting to quantify the frequency or probability of the event "10 green candles followed by 5 red candles", and the expected returns 30 days out. Another interesting query would be if it were possible to compute the probability of another red candle appearing, conditioned on 10 green candles in a row followed by 3 red candles in a row, i.e.,
p(R|[G,...G,R,R,R]).
Your post is very interesting and thought provoking. Thanks!
We can study a lot of patterns and ask what happens next or what is the likelihood of seeing one pattern followed by another one.
However there are limitations on that due to how small the sample size gets as soon as you start looking at complex patterns.
Here I'm working with daily candles so about 10 years of data isn't that much if we start searching for sequences of 10-15 candles.
One way to work around that is to do the same analysis using candles with higher frequency (like 4 hours, 1 hour and so on). But from my experience there is a point at which you end up getting a lot of noise.
That being said it would be interesting to see if there are "longer" patterns that arise at a higher frequency that you would expect. Maybe something I'll look at later.
Nick, when I switch timezone in TradingView, it doesn’t change how candlesticks look (the patterns). I tried with different intervals - still the same. So what did you exactly mean by that?
I'm not sure how TradingView works with respect to timezones.
But what I'm referring to is the following:
- Bitcoin does not have official trading hours like say the stock market.
- So depending on the data source you use, the daily close of a candle could simply be midnight of whichever timezone you are in.
- That means the close price would not be consistent across different timezones.
- As a consequence it would be possible to have the close higher than the open in one timezone but lower in another timezone resulting in a different candlestick pattern.
Apparently TradingView keeps the close fixed (see E. Hero's response) so as to have consistent candles whichever timezone you use. But that's not necessarily the case on all platforms.
It doesn't matter which time zone you prefer, as it will not adjust the candlestick closes. They are fixed to UTC (+8). This is more clear on the higher time-frame charts (daily, weekly, monthly).
You can't add the values together. You need to look at the expected return for conditional event "3 red candles knowing that the previous 10 candles were green". But the more complicated you make the pattern the less historical data points you get to estimate anything. So theoretically no and actually no.
Was being facetious, I understood your article and it was a good one might I add. It's the type of info-graphing I would attempt to make if I weren't so lazy. Shared with my group.
Dope!
I wonder if there has ever been a pattern of 10 green candles followed by 5 red candles before. This "10 green, 5 red" pattern seems relevant now: we just observed 10 green candles between 21-Aug to 30-Aug-2021 and we are now seeing 3 green candles in a row 31-July to 02-Aug. It would be interesting to quantify the frequency or probability of the event "10 green candles followed by 5 red candles", and the expected returns 30 days out. Another interesting query would be if it were possible to compute the probability of another red candle appearing, conditioned on 10 green candles in a row followed by 3 red candles in a row, i.e.,
p(R|[G,...G,R,R,R]).
Your post is very interesting and thought provoking. Thanks!
Thanks Sam!
We can study a lot of patterns and ask what happens next or what is the likelihood of seeing one pattern followed by another one.
However there are limitations on that due to how small the sample size gets as soon as you start looking at complex patterns.
Here I'm working with daily candles so about 10 years of data isn't that much if we start searching for sequences of 10-15 candles.
One way to work around that is to do the same analysis using candles with higher frequency (like 4 hours, 1 hour and so on). But from my experience there is a point at which you end up getting a lot of noise.
That being said it would be interesting to see if there are "longer" patterns that arise at a higher frequency that you would expect. Maybe something I'll look at later.
Cheers,
Nick
Nick, when I switch timezone in TradingView, it doesn’t change how candlesticks look (the patterns). I tried with different intervals - still the same. So what did you exactly mean by that?
I'm not sure how TradingView works with respect to timezones.
But what I'm referring to is the following:
- Bitcoin does not have official trading hours like say the stock market.
- So depending on the data source you use, the daily close of a candle could simply be midnight of whichever timezone you are in.
- That means the close price would not be consistent across different timezones.
- As a consequence it would be possible to have the close higher than the open in one timezone but lower in another timezone resulting in a different candlestick pattern.
Apparently TradingView keeps the close fixed (see E. Hero's response) so as to have consistent candles whichever timezone you use. But that's not necessarily the case on all platforms.
It doesn't matter which time zone you prefer, as it will not adjust the candlestick closes. They are fixed to UTC (+8). This is more clear on the higher time-frame charts (daily, weekly, monthly).
So if we end this 3 day red candle slip, we can add the two values together and expect a 90% return in 3 weeks? Theoretically yes, but actually no.
You can't add the values together. You need to look at the expected return for conditional event "3 red candles knowing that the previous 10 candles were green". But the more complicated you make the pattern the less historical data points you get to estimate anything. So theoretically no and actually no.
Was being facetious, I understood your article and it was a good one might I add. It's the type of info-graphing I would attempt to make if I weren't so lazy. Shared with my group.