Note: I needed to publish this with 15m timeframe. But for best results (more exact values) I always draw the lines with 1m timeframe, and scale up to 15m for evaluation.
The Theoretical Base: Mathematics.
The basis of my approach is the infinitesimal calculus - remember math in school? The first derivative (D1) of a function is the rise of its curve (here: how much a price rises or falls), the second derivative (D2) corresponds to the trend of the curve rise - i.e. whether a curve is concave (rise decreases, the curve makes a "hump" and tops out) or convex (rise increases, with positive rise perhaps parabolic phase).
The Road To Practice: Find It In The Chart
D1 corresponds here with the angle of a trend line, which I put tangentially to the curve. In contrast to the other procedure with rising curves from above, with falling curves from below. About D2 at this point we only need to know here whether it is positive (curve convex) or negative (curve concave). If we compare the angle of two adjacent tangents, D2 corresponds approximately to the difference of both angles: If the angle decreases (as in the chart), D2 is negative, the curve is concave.
Trade Point Definition
The perfect trade point for entering a short is the point where D1 has become most close to zero and D2 is negative - the top of the curve segment, its upper reversal point. From there on, it goes down. The best TP for the exit is the point where D1 is most close to zero with positive D2 - the lowest point of the curve segment, its lower reversal point. Note: In practice, local highs in a downtrend are broad. I have written about the "slow death" on tops before. Those are well suited for the application of the technique described in here. On the other hand, local lows in a downtrend tend to be as narrow as a singularity and hence are hard to catch. I will write another idea later on how to approach them most closely.
With this knowledge, I now try to find these two points by putting tangents to the price curve and comparing their slopes / angles. Of course, this can only be done approximately, because the price curve never runs smoothly like a sine, but oscillates in different timescales with different frequencies and these oscillations overlap to a chaotic oscillation.
So I define the timeframe in which I want to trade and ignore higher timeframes for the creation of the tangents. Then I create approximated tangents (trendline tool) on the outside of each curve section, which connect the high point of each subordinate timeframe with the next, without intersecting the curve at any point.