Dec

2

Here's a very educational article on why forecasts based on trend lines are always biased to look disturbing with some nice references, allusions, and diagrams from this list's favorite statistician, (also the namer of this humble old shaver's first child).

Gyve Bones adds: 

Trend lines, no matter how they are drawn, whether projecting rays into the future from price extremes, or drawing a least square fit line, or doing a polynomial curve fit line, or a moving average… One simply must understand how the line is calculated and take pains not to delude yourself by drawing a line into the past which contains information which would not have been available at that point in the past. A common way this is done is by drawing a trnedline through two lows, and extending that line leftwards from the first of the two lows. Most charting software does this by default, and it tricks the eye and the mind in a subtle way.

The problem with the original chart Larry posted was not the regression line per se, but in the lane below where an indicator was calculated showing how far above or below the regression line the price is. Here is where the contamination of the analysis happens. The indicator first calculates the regression line for the entire data set from alpha to omega. Then for every time interval between alpha and omega it measures where price is relative to that line and normalizes that value to some scale. then it looks at points in between where there was a high, like 1929 before the crash and says ah, the level of the indicator is 74, so anytime we get to a level of 74 we are in danger of a crash. But that number, 74 has encoded into it knowledge of the future. If one had calculated the same indicator in 1929, it would not have been 74 on that same date. The indicator is time-unstable.

The regression indicator posted in a subsequent post is time stable because it doesn't use information from the future, only information from the last to produce the value. This is a better way of using the tool, according to me.

I don't think we can generalize and say either trend lines are good or trend lines are bad. They can be used well or badly. He key is finding a way that will work logically and reasonably and that works for you. If you can make money using trend lines drawn on charts, then it is a good thing, subjectively. The only thing we can say objectively about the methods is that they can be dangerously deceptive if one doesn't understand how the lines are constructed, and one unconsciously is creating a tool that uses next week's Wall Street Journal as a source. That would be building a foundation for market analysis on sand.

anonymous adds: 

To the extent trendlines are useful the ideas surrounding how to "trade" them can be simplified. For example buying at a trend line support level is just a visual way of buying a dip in a rising market. There are simpler ways without the ambiguity to define this event that will most likely generate more events as well - making the underlying idea easier to evaluate. At the same time if one has evaluated millions of variations of a basic idea that tends to work or have some merit, what is the harm in trading that basic idea off of a visual cue? Perhaps the failure to scale well and movement toward mental laziness.

Sam Marx comments:

This can be said about many TA tools and indicators. For example, you cut a piece of data from alpha to omega, today is omega, calculate a mean and volatility for the sample, and then you may find that volatility at omega is extremely low which can lead to certain things. No one cries that somewhere before omega we measured volatility relative to mean which took into consideration the whole interval alpha-omega. All signals are usually generated for the period omega +1. I think the core weakness of these studies is that we can prove whatever we wish almost regardless of tools by simple selection of look back in these rolling derivatives of price.


Comments

Name

Email

Website

Speak your mind

3 Comments so far

  1. Travis Huff on December 2, 2014 5:34 pm

    Victor,

    Norman Lizt and I have a an interesting idea to discus with you (we may have spotted an enormous market anomaly).

    Please pass your telephone number on, we’ll call you.

    Travis
    415-409-8703

  2. armidal7 on December 3, 2014 8:02 pm

    yes. I sent it to you. you can contact me at lap@mantr.com always good to hear from NOrman after 55 years. vic

  3. Andrew Goodwin on December 5, 2014 2:54 am

    What the statistician traders miss is that the human brain is programmed to observe and create geometrical patterns. Since the observants admit to looking at price charts then they know that they might influence results.

    https://humanprehistory.wordpress.com/2014/12/03/homo-erectus-engraved-abstract-patterns-on-seashells/

    Homo erectus was making patterns on shells long before today’s speculators were trying to use patterns to turn a profit.

Archives

Resources & Links

Search