Very nice idea. Have you tried two and three candles? Also, an important aspect of candles is the context in which they appear. For example, a hammer in an uptrend. If clustering gets difficult, check out the mclust package which applies different clustering methods and number of clusters. It displays a chart with an overall goodness metric so you can choose a good starting point for number of clusters and method.
Statistical Analysis of Candlesticks patterns
- Thread starter AdrianHagh81
- Start date
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Exactly; uptrends [NasdaQQQ,SPY,DIA current trend friend]differ much from downtrends;
which differ much from sideways slop chop............................................................................................................
yeah im thinking of grouping one or more candles but it gets tricky. thx for thr info
ok, which period do u want to test for downtrend and choppy ?
I did some work in this direction last year. Worked with several ideas like calculating the high, low, close as a function of the open in order to push the candles into a 3 dimensional object, made for some nice and pretty charts.
Also played with ideas such as the relative size of the wick to the body, size of the upper vs lower wick, avg price of the candle (built from the ticks, and high+low/2 ... )
Tried different flavors of the k clustering algo, like manhatan distance, euclidean distance... etc...
Setting the initial centroids by hand took a little hacking of the libraries but came in handy to allow me to port results from one data set to the next without having to re-run everything every time.
I worked on these for about 3 months, but had to put them down since I wasn't getting any tangible results and had other more urgent needs (like getting my db up and running)
Also played with ideas such as the relative size of the wick to the body, size of the upper vs lower wick, avg price of the candle (built from the ticks, and high+low/2 ... )
Tried different flavors of the k clustering algo, like manhatan distance, euclidean distance... etc...
Setting the initial centroids by hand took a little hacking of the libraries but came in handy to allow me to port results from one data set to the next without having to re-run everything every time.
I worked on these for about 3 months, but had to put them down since I wasn't getting any tangible results and had other more urgent needs (like getting my db up and running)
I did some work in this direction last year. Worked with several ideas like calculating the high, low, close as a function of the open in order to push the candles into a 3 dimensional object, made for some nice and pretty charts.
Also played with ideas such as the relative size of the wick to the body, size of the upper vs lower wick, avg price of the candle (built from the ticks, and high+low/2 ... )
Tried different flavors of the k clustering algo, like manhatan distance, euclidean distance... etc...
Setting the initial centroids by hand took a little hacking of the libraries but came in handy to allow me to port results from one data set to the next without having to re-run everything every time.
I worked on these for about 3 months, but had to put them down since I wasn't getting any tangible results and had other more urgent needs (like getting my db up and running)
Well if you think there's nothing tangible out of it, gimme the code so I can continue where you left off
Redneck, why are you talking about "context"?
Candlesticks are what they are : a visual representation of price, IMO, the best one compared to the others (bars or linear graphs) because it gives you faster a clear global view.
And of course candlesticks patterns help to give probabilities of upside, downside or sideways scenarios in a ST, MT and LT perspective. But i agree, we are talking about probabilities and not forecasting.
If u candle guys can quantitatively define what you mean by context and etc i can run the sim to test. Otherwise
It'll be too vauge to gauge.
It'll be too vauge to gauge.
Can you elaborate on the differences between probabilities and forecasts?