Quote from: JorgeStolfi on November 13, 2014, 01:40:41 PM

Quote from: oda.krell on November 13, 2014, 12:22:31 PM

one day, when you're ready, you will see the (technical) light, and identify the two main drivers of markets (outside fundamentals): momentum, and mean reversion Cheesy

Seriously, here is one of the reasons why I can't take most TA seriously:

An n-day moving average (MA_n) is the average of n consecutive prices. Charting the moving average instead of the price reduces the random day-to-day noise while preserving the longer-term variations.

An exponential moving average (EMA) is similar, but gives more weight to the recent behavior -- and is easier to compute if you are using pencil and paper or a slide rule. 8-)

The difference between two MAs with different periods -- say, 7 and 21 days -- is basically a smoothed version of the second derivative of the price. If MA₇ is higher than MA₂₁, the trend of the price is bearish, curving downwards: it is braking while going up, then accelerating down. Conversely, if MA₇ is lower than MA₂₁, it means that the trend is bullish, curving upwards: braking while going down, then accelerating up. When the two graphs cross, then, the trend is changing from one mood to the other.

Good so far. However, the average of the last 21 prices tells what the price was 10 days ago, not today. But analysts always plot the MA₂₁ value at the final date of the 21-day interval, not at the central date. So, the traditional MA₂₁ plot is displaced 10 days to the right relative to its logically correct position (and you can see that clearly when it is plotted over the daily price chart). Similarly, the MA₇ chart is shifted 3 days to the right. Therefore, the two plots cross at the wrong dates; if they were drawn at the correct dates, crossings might appear or disappear.

TA folks even understand this problem, but they have been plotting their MAs and EMAs that way since the lower Paleolithic, and cannot be convinced to change their habits. There may also be a psychological factor: plotting the MA_n values at the correct dates makes it obvious that the analysis is based on information that is 10 days old; whereas plotting it the traditional way gives the impression that the analysis is hot from the ticker's mouth.

There is also the problem that the crossings depend on the arbitrary choice of the MA periods. It is like computing the ROI of bitcoin investment: one gets completely different results depending on the time scale considered.

Finally, this MA-crossing analysis depends on the hypothesis that there is an underlying price trend that can be revealed by smoothing the price. That may be true in ordinary stocks, where the price is partly determined by fundamentals that change slowly with time (such as the overall economy and product sales). It is rather questionable with bitcoin, since its "fudamentals" are isolated unpredictable events (like PBoC decrees and rumors). We see that in the charts, where rallies and crashes often start suddenly, out of nowhere.

It doesn't matter what TA events are or how the lines that indicate them are formed so much as it matter how many traders react to those events. I could have two random lines on the chart, but if 90% of traders are going to trade their intersections, then it's still a valid indicator, a self fulfilling prophesy. TA doesn't have to predict the future, if it causes the future.