Sunday, November 15, 2009

New wave study

It's Sunday morning and I have some time playing around with an idea I already had for quite some time.

Prices move in waves. I think we can agree on that and using charts means, we try to predict future price behavior from past behavior aka prices are not moving randomly.

Also looking at any line chart in any timeframe, we see waves within waves. And usually we see breakouts (up or down) followed by consolidation followed by a new wave up or down and again consolidation. We use divergence, price patterns, whatever to tell us, it's time to take a trade, because we believe to see a pattern we recognize and which tells us, that a certain move is imminent.

clip_image002 
(Friday's 5min ES chart European time zone)

I thought of a theoretical way to get a similar chart and a simple formula was able to do that:

y=sin(x)

gives you a standard sinus wave which in it's simple form might represent buying, consolidation at the top, selling, consolidation at the bottom and an endless repetition of the cycle

Math4

y=sin(x) doesn't do the trick, but if you define x to be different distinct timeframes, each a timeframe with a group of traders working that timeframe and you just add them up, like each wave in the ocean is just the sum of all waves it is composed of, you get a very interesting picture:

clip_image004

Gives you this graph:

clip_image006

In this graph you see a lot of patterns, we use for our daily trading.

And I thought to myself, if there was a way to match the line chart of the ES above to such a theoretical chart, you might get a really good predictive study.

I have no idea, how such a predictive wave study could be implemented, but maybe it's an idea worth thinking more about. Maybe you don't need a complex neural net to try to explain market behavior within the day, maybe a relative simple Wave study is sufficient. You need to play with the parameters and the number of Waves within waves, to get a good representation of the market, but starting with the major timeframes should work. Volatility could be added by multiplying certain timeframes to match news events pre-market.

clip_image008

It's just an idea, but maybe you have some time to follow up on it and play around with it.

5 comments:

Anonymous said...

Hi Chris,

Your post raises what seems to be a very interesting idea. I would love to be able to add to the discussion but sadly my maths is far too basic for these thoughts of yours.

All the best,

David_uk

ART said...

Hi

You are trying to find the underlying frequescies and the apropirate amplitudes of a timeseries you have to use Fast Fourier Analyse Algo to do this.

The Problem is you will get the different frequencies and the the aplitudes for the past data, but this amplitudes are not stationary and you have to aproximate the new values for the future.

greetings
ART

leon t said...

chris: you may want to look up leonard novy, he already has a similar concept going. you'll find him amusing and helpful.

Globetrader said...

So Art, what you are actually saying is, even if I get a good FFA the past still won't predict the future.
...
You might be right and I might need to think along other lines.
Best regards,

Chris

Anshuman said...

Chris, you ought to look at Didier Sornette's book on stock market crashes. The patterns he models are timescale-invariant, and might be of interest to you.