i posted this in the new hardcore thread: https://bt.irlbtc.com/view/178336.msg1862366#msg1862366

thank's for drawing the lines! although, the case hasn't been quite resolved yet … we'll see. this could also be the beginning of an upside down cup, like before.

My idea:



We are not going up. Not yet.

The price was being held down under the down channel upper wall very well. In your chart the price isn't actually hitting the upper line at all for most of the last week. I think your upper line needs to be lower, in which case the price just broke through it.

30-min PVT says something big is brewing...

-===-



-===-

this is the decisive moment... the only question is -- up or down?

to all the newbies drawing trendlines and triangles -- just some tips:

1. you need at least two points of contact. a line is defined by two points so you can draw any line you want about one (this is not rigorous enough, obviously). further, if your line only has two points of contact, it is not as robust as a model that has more.

2. sometimes wicks (shadows) contain outliers. try to imagine the distribution of the data and draw lines accordingly -- it's TA, not connect-the-dots. beware of apophenia. just because you see a pattern doesn't mean it's a proper, predictive, or particularly robust model.

… and if you know a bit about models and fitting: i see this as an approximation problem. e.g. take all the max-points for a given sliding and overlapping interval (i.e. "top points") and approximate them via a linear function. (it's also linear if you take the logarithm). then, the residual could be calculated using the L-1 or L-2 norm. L-1 referrs to point 1 of arepo, L-2 is his/her second point.
Advanced data fitters might also consider to assign weights to the points, e.g. the last points have an exponentially higher weight than the first ones. e.g. in python/numpy: np.exp(np.linspace(-1, 0, 10)) if you have 10 points.
in python's "statsmodesl" it's OLS and WLS.
… and if you have more points to fit and want to take outliers into account, look into RLS: http://statsmodels.sourceforge.net/devel/examples/generated/example_rlm.html