When Technical Analysis is Not Appropriate for Algorithmic Trading (Part 1)

While technical analysis (TA) can be useful, not all TA approaches are amenable to algorithmic trading.

Elliott Wave Theory (EWT) is one such approach that identifies patterns created by impulse waves and corrective waves that oppose the larger trend. This is a fractal approach because each set of waves is nested in a larger set of waves that follow the same impulse-corrective pattern. The problem with EWT is the use of alternate wave labels that remain until the pattern completes. Only the final (correct) labeling remains on the chart. It therefore looks perfect in hindsight every time.

I was once told for every two EWT traders who give opinions, three possible wave profiles will be shown. One review of Advanced Get (EWT software) claims “it always looks great after the move but as the move progresses the software changes the count.”

If we could only trade real accounts this way! optionScam.com anyone?

Distilled into faults, EWT is a form of TA that makes up rules as it goes along, looks perfect in hindsight, and cannot be programmed because the guidelines are equivocal. If the latter were not true, then alternate wave profiles would never appear.

Hindsight bias also pertains to trendlines. Trendlines vary depending on things like time frame, time interval of the whole chart, how many touches the interpreter requires to be valid, and how exact the touches must be. This is not an exhaustive list. Similar to the comment three paragraphs above about EWT, I’ve heard it said many times that you will get x different sets of trendlines drawn for every x traders you gather in a room.

Computers may be programmed to calculate trendlines by matching open/high/low/close (OHLC) values, which are commonly included in historical price databases. We know that one and only one line is defined by any two points. What if three or more collinear OHLC points are required, however (do an internet search for “trendlines require three touches”)? It may be the case that in the last n bars, no lines contain three or more O, H, L, or C’s. A margin of error could be built-in to allow three or more nearly collinear points, but this would likely result in multiple sets of trendlines and a really messed up trading strategy.*

I will continue next time.

* — It might be useful at a later point to explain in detail why I say this is “messed up.”