On why traditional tech analysis stopped working and on when it shall start again...

[sinner]

http://www.priceactionlab.com/Blog/2015/05/fooled-by-persisting-market-conditions/

Many classical chart patterns worked well in the 1970s and 1980s due to the high serial correlation. The shape of the patterns was mostly irrelevant although some behaved better than others due to their duration. The key behind their performance was that there was high probability that up days would be followed by up days and down days by down days. As a result, a double top or a head and shoulders worked as long as they were confirmed. But that is not the case any longer and false breakouts prevail. Thus, chartists of the 1970s and 1980s were fooled by high serial correlation. When the serial correctional became negative, charting along with most indicators stopped working. The market was ruthless in that respect and took the money of naive technical traders. Some still struggle with such methods, not doing their homework. Some others hope that serial correlation will return to the markets before they are too old to trade. I hope it will because there should be some form of wealth redistribution among traders. But the situation is much more complicated and instead of a return to high serial correlation we may see something else.

I endorse this analysis.

This also means that if you can find assets with high serial correlation, traditional tech analysis should work well. I normally use tools like the TSI to find assets of this variety.

http://www.nonrandomwalk.com/Blog/tsi.html

https://engineeringreturns.wordpress.com/tsi/

(and it works in the opposite function too, low TSI score assets have low/negative serial correlation and generally make for good mean reversion trades).

 

[tech/a]

Re: On why traditional tech analysis stopped working and on when it shall start again

TSI ?

 

[peter2]

Re: On why traditional tech analysis stopped working and on when it shall start again

TSI = Trend Strength Indicator

TSI is described in the second reference.

Thanks sinner, you've mentioned the work of David Varadi before and you've rekindled my interest in his work again.

 

[tech/a]

Re: On why traditional tech analysis stopped working and on when it shall start again

I see this as nothing more than saying T/A works better when trading with a trend.

Rising tides float boats.

Changes in conditions particularly a change relative to past price action and your system could fail.

TSI is a technical filter.

How it helps keep a system on the right side only you know.
Unless you have some tests to show us?

Perhaps the adaptation of T/A hasn't kept up with market changes (Particularly in his case)
But as the DJIA in the 70s and 80s Basically went side ways at around 1000
and from then till now went 1000-18000 pretty well vertical

I cant see where he's coming from.
Any proof of concept?

Click to expand

 

[tech/a]

Re: On why traditional tech analysis stopped working and on when it shall start again

On TSI.

I see this as just another lagging oscillator.
Used as a filter as suggested above 1.65 being the value
range in which T/A is likely to "Work" only has meaning if it stays above
1.65.

Practical application
Putting the idea to work.

Shown is the chart offered up as evidence of how it works.
About half the trend is captured.

But put the oscillator on a search and you'll find at least 50% of prospects that fail.(Say after 6 mths)
So the question arises
How do you apply it profitably to trading technically. Few are going to continue above or below 1.65 as seen on the chart offered.

Click to Expand



My argument is that this indicator is no better than any other as a filter---you could use an M/A,RSI,Stochastic.
Bollinger band,Trend line.

After 20 yrs of trading T/A my view is that its as valid today as it was when it was first discovered---whatever form of T/A or F/A one adopts.

What is also constant is the number of exponents of ANY ANALYSIS who know how to apply it profitably.
I've seen some really bad application of T/A on this forum and some really bad application of F/A on the same forum.

I argue its about the applier of analysis not market correlation. (Which is a simple cop out that anything works in a screaming Bull/Bear market.---actually many seem to be able to fail dismally here as well!)

 

[DeepState]

Re: On why traditional tech analysis stopped working and on when it shall start again

My

What is required for this to be useful is for the TSI to be highly persistent beyond the lag effects inherent in its construction. If high TSI today predicts high TSI in 100 days from now, that's when you can have confidence that application of trend following techniques can be profitable in the series. If low TSI can be expected to persist, then you can be somewhat confident that reversionary methods are likely to be profitable.

Is there any suggestion that the TSI is sticky beyond lag effects? If so, is this stickiness regime dependent (eg. stickiness might be expected if the series were cheap and getting more expensive as value and momentum considerations create a strong trend, but not over a full cycle on average). If so, then you ultimately need to know the regime...

In line with T/A, I can agree that it is an historical observation of trending behavior. For it to be useful for money making, it also needs to be a predictor. Is it?

 

[sinner]

Re: On why traditional tech analysis stopped working and on when it shall start again

I see this as nothing more than saying T/A works better when trading with a trend.

Rising tides float boats.

Changes in conditions particularly a change relative to past price action and your system could fail.

TSI is a technical filter.

How it helps keep a system on the right side only you know.
Unless you have some tests to show us?

Perhaps the adaptation of T/A hasn't kept up with market changes (Particularly in his case)
But as the DJIA in the 70s and 80s Basically went side ways at around 1000
and from then till now went 1000-18000 pretty well vertical

I cant see where he's coming from.
Any proof of concept?

Click to expand

View attachment 62729

*sigh*

Yes tech, TSI is a technical filter. No, it is not like an MA, or stoch or Bollinger or whatever. It is simply an average of the measure of absolute returns divided by an approximation of volatility. It doesn't help to keep you on the right side of a trade. Its job is to help keep you in the right kind of trade. You have stated yourself that you manage your portfolio in such a way that low trend participation stocks are sold for stocks with better trend participation. This is simply an attempt to quantify that. You can also simply use serial correlation.

The fact that you can't see where he is coming from does not surprise me, since it's obvious you didn't bother to read anything I posted. Yes, there are lots of tests, done by the inventor (Frank Hassler) using the tool in various things from short term index mean version to using it as a ranking component in relative strength systems. You can see the results of those tests in the link I already posted..

David Varadi also has extensive tests on his blog.

I see this as just another lagging oscillator.
Used as a filter as suggested above 1.65 being the value
range in which T/A is likely to "Work" only has meaning if it stays above
1.65.

The actual threshold which defines "trending" versus "not trending" is different depending on the asset class. The implication of "trending" state is simply that serial correlation is higher (i.e. up days are proceeded by up days) and implication of "not trending" state is that serial correlation is lower (i.e. up days are proceeded by down days).

How you make use of that is really up to you, but I personally find it useful in conjunction with traditional trend following to break the "regime" of the asset in question down into four states:

Uptrend with no participation
Uptrend with good participation
Downtrend with no participation
Downtrend with good participation

Can you guess which regimes breakout systems work best in? And which regimes have the highest probability of breakout failure? Do you get that breakout success is due to high serial correlation and that breakout failure is due to low serial correlation?

You can't do this quantitatively (at least not as simply) with only trend following tools.

Practical application
Putting the idea to work.

As mentioned above, in the links I provided (and via a tiny tiny bit of googling) you could easily find smart people who have used this tool as a component in trading systems which have holding times ranging from 1 day to 31 days.

Shown is the chart offered up as evidence of how it works.
About half the trend is captured.

Again, the point is not to capture the trend.

But put the oscillator on a search and you'll find at least 50% of prospects that fail.(Say after 6 mths)
So the question arises
How do you apply it profitably to trading technically. Few are going to continue above or below 1.65 as seen on the chart offered.

Very, very easily.

My argument is that this indicator is no better than any other as a filter---you could use an M/A,RSI,Stochastic.
Bollinger band,Trend line.

What a stupid argument.

I argue its about the applier of analysis not market correlation.

Did you actually even bother to attempt to understand what serial correlation means? It doesn't seem so. Maybe ask your son.

 

[sinner]

Re: On why traditional tech analysis stopped working and on when it shall start again

My

What is required for this to be useful is for the TSI to be highly persistent beyond the lag effects inherent in its construction. If high TSI today predicts high TSI in 100 days from now, that's when you can have confidence that application of trend following techniques can be profitable in the series. If low TSI can be expected to persist, then you can be somewhat confident that reversionary methods are likely to be profitable.

The persistence is present, but due to nothing more than momentum effect and inherent predictability of vol, certainly not due to any magical property of the indicator or magic of its construction. After discovering the TSI I made several indicators which do essentially the same thing but use different inputs (e.g. historical vol rather than ATR) and have observed the same result across markets.

 

[skyQuake]

Re: On why traditional tech analysis stopped working and on when it shall start again

Great stuff gents.

I just started reading this recently too:

HURST EXPONENT AND FINANCIAL MARKET PREDICTABILITY

ABSTRACT
The Hurst exponent (H) is a statistical measure used to
classify time series. H=0.5 indicates a random series
while H>0.5 indicates a trend reinforcing series. The
larger the H value is, the stronger trend. In this paper we
investigate the use of the Hurst exponent to classify series
of financial data representing different periods of time.
Experiments with backpropagation Neural Networks
show that series with large Hurst exponent can be
predicted more accurately than those series with H value
close to 0.50. Thus Hurst exponent provides a measure
for predictability

 

[sinner]

Re: On why traditional tech analysis stopped working and on when it shall start again

Great stuff gents.

I just started reading this recently too:

HURST EXPONENT AND FINANCIAL MARKET PREDICTABILITY

From my notes:

http://www.bearcave.com/misl/misl_tech/wavelets/hurst/

 

[sinner]

Re: On why traditional tech analysis stopped working and on when it shall start again

My personal experience with Hurst cycles was not great in various markets. I generally found that (like many other things) that simplicity lends to robust results and complexity less so.

So generally speaking, my findings have been that simpler measures (%rank of serial correlation, various measures of absolute returns divided by various measures of volatility, etc) have been much more effective.

 

[CanOz]

Re: On why traditional tech analysis stopped working and on when it shall start again

Looks about right....

 

[CanOz]

Re: On why traditional tech analysis stopped working and on when it shall start again

Here is the TSI with one of the strongest 'trends' in the world right now, the Shanghai Composite.

 

[tech/a]

Re: On why traditional tech analysis stopped working and on when it shall start again

*sigh*

I am indeed honoured to have received your learned reply.

 

[sinner]

Re: On why traditional tech analysis stopped working and on when it shall start again

Here is the TSI with one of the strongest 'trends' in the world right now, the Shanghai Composite.

I didn't really want to get the thread hung up on TSI, the point was really more about what is actually driving the returns (or lack thereof) in traditional technical analysis setups.

But anyway, as mentioned above, I wouldn't recommend arbitrarily applying 1.65 as a useful threshold across different asset classes (e.g. individual stocks are quite different to stock indices, which again very different to commodities).

My suggestion is to use a long term median value as the threshold, e.g. ~500 day avg of TSI or watch a shorter term ROC to see when the TSI starts rising and falling.

 

[lftrader]

Re: On why traditional tech analysis stopped working and on when it shall start again

http://www.priceactionlab.com/Blog/2015/05/fooled-by-persisting-market-conditions/

I endorse this analysis.

This also means that if you can find assets with high serial correlation, traditional tech analysis should work well. I normally use tools like the TSI to find assets of this variety.

http://www.nonrandomwalk.com/Blog/tsi.html

https://engineeringreturns.wordpress.com/tsi/

(and it works in the opposite function too, low TSI score assets have low/negative serial correlation and generally make for good mean reversion trades).

Thanks, this is interesting. On another note, has anyone tried the Price Action Lab software? That also looks interesting to me but more as a scan tool.

Another question, is this similar to the TSI?

 

[sinner]

Re: On why traditional tech analysis stopped working and on when it shall start again

Thanks, this is interesting. On another note, has anyone tried the Price Action Lab software? That also looks interesting to me but more as a scan tool.

Another question, is this similar to the TSI?

Kind of, sort of, but not really.

When the market is “stable” it is easier to apply effective quantitative trading systems. When the market is in “chaos” mode, it is not necessarily volatile- but rather it is too complex to use for standard measurement and calibration of basic linear prediction.

If you think of the TSI as an example of an indicator which tells you whether to use your trend following or mean reversion system, the self similarity metric is an example of an indicator which tells you whether or not to use any system.

I've never tested it so can't say how effective or not it is at this. However I am aware of other indicators which are used similarly as filters for machine learning based systems (not sure whether or not they are capturing the same phenomenon though).

 

[tech/a]

Re: On why traditional tech analysis stopped working and on when it shall start again

Sinner.

Had a chance to have a look at all the thread references you posted up with regard to the heading of your thread.

I think there is a very long bow drawn with regard to "proof" that Traditional Technical Analysis has stopped working.
Offered up are two rudimentary systems the RSI (2) and a Two Day losing streak.(Hardly encompassing All of the available " Traditional Technical Analysis" available).
Then benchmarks it against the S&P 500 and runs to two methods in the two timeframes showing that high correlation resulted in lower returns on these two tests against this benchmark.

To then conclude that all Chartists were fooled by high correlation in the years selected and that traditional Technical analysis has stopped working--I think is a very very broad brush.

TSI is still in my opinion another filter.
How it helps needs far more testing against a greater range of bench mark/systems/analysis and data.

I think its a long way from proving that traditional Technical Analysis has stopped working.

But worth investigation.

I also note that his software is often referenced.

 

[Habakkuk]

Re: On why traditional tech analysis stopped working and on when it shall start again

Michael Harris is an interesting guy. Many of his posts make perfect sense (to me anyway) although their main purpose seems to be to promote his software at every opportunity. Can't blame him, really - it's his business. But then you look at what his Price Action Lab software does. I won't go into it here; you can look it up or you can download a (crippled) trial version to get an idea, which I have done.

Now, in my humble non-expert opinion it has little or no value, not even 50 bucks. No explanation for now, just my opinion.
However, I would not be prepared to bet even money against one of its signals that has shown a 90% success rate on backtests. Hard to explain that logic, so I won't.
But, like lftrader, I would like to hear what the more knowledgeable members think about the Price Action Lab concept. I don't believe that anyone here has forked out the US$ 2795 he charges (used to be 5 grand). But could you just read the description and give your considered opinion?

http://www.priceactionlab.com/index.html

I said he was an interesting guy. You'd reckon with that software that is so valuable, and all his happy subscribers, he could afford to upgrade from Windows XP. Apparently not.

 

[sinner]

Re: On why traditional tech analysis stopped working and on when it shall start again

Sinner.

Had a chance to have a look at all the thread references you posted up with regard to the heading of your thread.

I think there is a very long bow drawn with regard to "proof" that Traditional Technical Analysis has stopped working.
Offered up are two rudimentary systems the RSI (2) and a Two Day losing streak.(Hardly encompassing All of the available " Traditional Technical Analysis" available).
Then benchmarks it against the S&P 500 and runs to two methods in the two timeframes showing that high correlation resulted in lower returns on these two tests against this benchmark.

To then conclude that all Chartists were fooled by high correlation in the years selected and that traditional Technical analysis has stopped working--I think is a very very broad brush.

TSI is still in my opinion another filter.
How it helps needs far more testing against a greater range of bench mark/systems/analysis and data.

I think its a long way from proving that traditional Technical Analysis has stopped working.

But worth investigation.

I also note that his software is often referenced.

You're taking it personally, because you think when he says "technical analysis" he's referring to what you do. He is not. He is talking about why, for example, setups like "head and shoulders", or indicators like MACD, used to work when they were invented but no longer do.