Analysis of hanging man

[iateadonut]

hi,

i started running simulations to see the likelihood of certain stock patterns having an indication on future price.

i got a book on japanese candlesticks, where the guy purports that hangman is quite the accurate indicator.

i ran an experiment. the details are below. the link is of a pdf of 100s of graphs with hangmen. i'm wondering if anyone can tell me why i'm getting a 50% success rate (like flipping a coin).

also, if anyone feels that there are more accurate indicators that i should be studying, i'd be interested to hear what they are.

(forum is not allowing me to post my pdf)

An experiment was done to find hanging men in 2010, as defined above (shadow 2.5x length of stick or more; candle at top; almost no top shadow). After the hanging men were found, the trend before the hanging man (hanging man close minus(-) close of 5 days before) was compared to the trend after (five days after minus hanging man close); only the signs (+ or -) of the two trends were compared. All stocks listed on the NYSE were compared.

It was found that the likelihood of a trend reversal following a hanging man was about 50%. In other words, under these conditions, a hanging man was about as reliable as tossing a coin.

Figures can be found in the spreadsheet on sheet '001'.


Suggestions for further study:
1) Test if the color of the candlestick provides any more reliability. (Although the real body of the hammer or hanging man can be white or black, it is slightly more bullish if the real body of the hammer is white, and slightly more bearish if the real body of the hanging man is black. "Japanese Candlestick Charting Techniques" Steve Nilson, p.29) So, if the initial trend is down, close>open. If the initial trend is up, close<open.

2) Test the strength of the trends. For example, any preceding (since the goal is forecasting) trend less than a 2% change or more than a 5% change might be thrown out.

3) Lengthen the lower shadow.


EXPERIMENT 2.

repeated with a longer tail (3.5x). results were the same.



EXPERIMENT 3.

Only hangmen with bullish candles were counted as trend reversals for downtrends, vice versa.
about the same as a coin toss.

EXPERIMENT 4.

test strength of prior trend. (2%-5%).


EXPERIMENT 5.
repeat criteria in 3 and 4 and extended tail (5.5x) and change strength test (1.9%-3%).
still may as well be flipping a coin.


FURTHER EXPERIMENTATION:
confirmation - will a open price lower (on the day after) than the close of the hangman day tend to confirm a price 5 days later?

"the general principle for the hanging man;
the greater the down gap between the real body of the hanging-man day and the opening the next day the more likely the hanging man will be a top.
Another bearish verification could be a black real body session with a lower close than the hanging-man sessions close."


EXPERIMENT 006.
no strength test. tail=3.5x. only for bearish trend reversals. does a lower open on the second day confirm the hangman?

EXPERIMENT 007.
same criteria as 006, but lower open of 97% or less of the close.



The previous experiments indicate that a hangman is about as reliable as a coin flip for determining the direction of 5 day trends.



EXPERIMENT 008.

100 random stocks were picked from the NYSE and analyzed for 2010(year) hangmen. All were charted to see if a human being might detect some pattern.

 

[tech/a]

I'm not suprised by your results.
Add the following filters and see how it goes.
For both bullish and bearish.

Volume over the past 2 days (any of them) or the hang man itself
3-6 or more times the average volume over the last 5 bars (Test in steps 3x,4x,--etc) So one bar of the last 3 must have a high increase in volume.

Volume of the FIRST up or Down bar from the signal lower than the signal bar.

So the key is seeing some exhaustive long or short volume and then clear movement in the opposite direction without the impedement of excessive volume (This is just as important as high volume--infact more so).
If you doo see high volume AFTER the hanging man then chances are the reversal isnt quite ready yet.

Let me know how you go.

This is an example (Not with candlesticks but "similar") you'll see the same in your test cases.
This is last nights FTSE 6 min bars.

 

[Gringotts Bank]

Bulkowski at the pattern site has done all this for you. You can see all the stats there, plus some trading set ups.


http://thepatternsite.com/CandlePerformers.html


In summary, they're not very useful. Patterns are more reliable, but harder to find and almost impossible to backtest. He often counts patterns manually when "backtesting" and that must be very hard work.

 

[pixel]

Bulkowski at the pattern site has done all this for you. You can see all the stats there, plus some trading set ups.


http://thepatternsite.com/CandlePerformers.html


In summary, they're not very useful. Patterns are more reliable, but harder to find and almost impossible to backtest. He often counts patterns manually when "backtesting" and that must be very hard work.

Thanks for the Bulkowski link, GB - bookmarked it right away.
I'm always looking up LeavittBrothers, whose explanations I find very concise and reliable too. But this guy claims to have analysed Millions of patterns and gives exact statistics. Amazing!

 

[Timmy]

I'm not suprised by your results.
Add the following filters and see how it goes.
For both bullish and bearish.

Volume over the past 2 days (any of them) or the hang man itself
3-6 or more times the average volume over the last 5 bars (Test in steps 3x,4x,--etc) So one bar of the last 3 must have a high increase in volume.

Volume of the FIRST up or Down bar from the signal lower than the signal bar.

So the key is seeing some exhaustive long or short volume and then clear movement in the opposite direction without the impedement of excessive volume (This is just as important as high volume--infact more so).
If you doo see high volume AFTER the hanging man then chances are the reversal isnt quite ready yet.

Let me know how you go.

This is an example (Not with candlesticks but "similar") you'll see the same in your test cases.
This is last nights FTSE 6 min bars.

Really great stuff tech.

 

[nomore4s]

i'm wondering if anyone can tell me why i'm getting a 50% success rate (like flipping a coin).

Because you need to add some context to the set ups. Look at the whole story not just a page or chapter.

As Tech has said you need to look at them as more of a multi bar set up, as well as taking other factors like general market conditions and trends into account.

IMO one of the secrets of trading that is rarely discussed is the ability to see the market as a whole and understand the bigger picture - ie - market trends, seasonality factors, the ebb & flow of money and risk (where money is running to or from - where risks are being taken or avoided) etc etc.

 

[iateadonut]

Volume over the past 2 days (any of them) or the hang man itself
3-6 or more times the average volume over the last 5 bars (Test in steps 3x,4x,--etc) So one bar of the last 3 must have a high increase in volume.

So the key is seeing some exhaustive long or short volume and then clear movement in the opposite direction without the impedement of excessive volume (This is just as important as high volume--infact more so).
If you doo see high volume AFTER the hanging man then chances are the reversal isnt quite ready yet.

Just to clarify, if we are looking at 20 days, day10 being the day of the hangman, we should see:
da7-day9 - any day should have volume greater than x (i'll start with 1.5x) times than the average volume of day1-6.

day11's volume should be < average(day1-6)

your suggestion is that i proceed in this direction?

 

[iateadonut]

Bulkowski at the pattern site has done all this for you. You can see all the stats there, plus some trading set ups.

In summary, they're not very useful. Patterns are more reliable, but harder to find and almost impossible to backtest. He often counts patterns manually when "backtesting" and that must be very hard work.

Thanks. I'll check that out an probably buy his book. I'm not the trusting type, so naturally I'll have to run my own experiments and confirm his. Maybe six years from now when I'm 36 I can also retire.

What do you mean "backtest"?

I have no desire to do anything manually, as humans are prone to see what they want to see. Only numbers for me.

 

[waza1960]

Ninja trader also has some basic candlestick patterns which you can backtest on etc which is quick and convenient IMO

 

[Timmy]

Because you need to add some context to the set ups. Look at the whole story not just a page or chapter.

As Tech has said you need to look at them as more of a multi bar set up, as well as taking other factors like general market conditions and trends into account.

IMO one of the secrets of trading that is rarely discussed is the ability to see the market as a whole and understand the bigger picture - ie - market trends, seasonality factors, the ebb & flow of money and risk (where money is running to or from - where risks are being taken or avoided) etc etc.

Great stuff 4s.

Steve Nison did a good job with his candlestick books in bringing the candlestick idea to the assessment of price action. But I think readers of his books do themselves a disservice in viewing single candlestick bars out of context of existing price/volume activity. I don't think candlesticks were ever meant to be viewed as single, isolated events.

Good thread iateadonut.

 

[tech/a]

Just to clarify, if we are looking at 20 days, day10 being the day of the hangman, we should see:
da7-day9 - any day should have volume greater than x (i'll start with 1.5x) times than the average volume of day1-6.

Yes

Dont complicate yourself,This science isnt exacting.(Try Elliott Wave and you'll know what I mean).

day11's volume should be < average(day1-6)

your suggestion is that i proceed in this direction?

Less tha the day of the hangman
so if it was 10 million and you have 1 then thats great.
It must also finish above its open.
Total confirmation is trading above the high of the signal bar or the low whatever the case maybe.

No more 4s
comments are spot on.

I'm sure youll find less setups but an edge.
Interested in your results.

 

[iateadonut]

Yes

Dont complicate yourself,This science isnt exacting.(Try Elliott Wave and you'll know what I mean).



Less tha the day of the hangman
so if it was 10 million and you have 1 then thats great.
It must also finish above its open.
Total confirmation is trading above the high of the signal bar or the low whatever the case maybe.

No more 4s
comments are spot on.

I'm sure youll find less setups but an edge.
Interested in your results.

One last question. How do we determine an uptrend/downtrend?

For example, would we say that 10% gain in 30 days is an uptrend or what?

(I realize this is painstaking, but computers don't just look at graphs and say, "look at the uptrend.")

Thanks a lot for all the help.

 

[professor_frink]

For a short term uptrend, you could look for today's close to be in the top 10% of the range for the past 20 sessions, for a longer term trend, 50 or 100 days(or whatever you think will capture the type of move you are looking for leading up to the signal). If you look for a set % move up, what you'll find is that during low volatility periods, you might find your % filter too large and not trigger, and in a higher vol environment, it'll be too small and will trigger after only a couple of up days.

 

[tech/a]

One last question. How do we determine an uptrend/downtrend?

For example, would we say that 10% gain in 30 days is an uptrend or what?

(I realize this is painstaking, but computers don't just look at graphs and say, "look at the uptrend.")

Thanks a lot for all the help.

As prof points out there are many ways
By eyeball i prefer higher highs lower lows.

I use a linier regression formula (Ma of regression) for computer determination.
But a simple M/A is OK say above or below a 21 period Exponential for shorter term
63 for longer.Try both

In the chart below Green is my Linier regression M/A
and the Blue is a 21 day ema.
Chart is AXO
21/05/10 looks interesting!

CLICK TO EXPAND

 

[tech/a]

Any further developements?

 

[iateadonut]

not yet. just got back from seoul, moving to FL in a few days so won't have time for a few days i think.