Technical Analysis alone will not give you an edge

I think what AlphaOmega is trying to get at is this idea that there are certain "moments" (yes I like to think of them as moments) in the market where it offers fantastic opportunities for you to get really really big without having to take on significant risk. These are the points where it is JUST BEFORE a huge move happens. If after adding size, the move does not materialise, then you know it's time to bail. It can be within 60 seconds of adding to the position. These are the moments that, if you learn how to identify, you can put up to 5 times your usual size and make huge profits.

Mastering trading is not just about trading in the right direction of the move. If you can take it further by timing the move correctly, you will be tremendously profitable.

price is not random, that is why technical analysis gives me an edge.

Nice! I've seen this kind of situations a couple of times, but I'm still keep on trying to find out how I was able to ride on it successfully.

Thank you alphaomega!

As far as my experience goes, when attempting that on my daily trading routines, my profitability always dropped -70% (i think, kinda forget ) even if i get the direction right that day.
Many things just doesnt sit well with my trading principles, one of the ex. is many times i get into this consecutive losses and after 3 or more i just cant afford to lose any longer, so i trail my stop into positive territory, price retrace then hit my stop. most of the time this situation gives me practically a (lucky) BE trade, as if i never trade that day .
But there are this rare times when the momentum is way so strong that my first/second attempt shoot right into to my intended direction.
So i realize that the momentum is indeed important, so to test this out i try combination of renko (size 2) + High impact news event.
When testing this out, i always assuming that the first burst is the intended direction, so if its go UP i wait for the retrace, and wait for crazy spread fluctuation to be normal again (sometimes only for few seconds, sometimes it takes forever ),.. renko make V shape and enter.
given the nature of renko, SL is always at fixed number (3), depends on the brick size.
Becos of strong momentum i find that i rarely makes re-entry attempt. From my entry point usually price just moves straight either up or down.
So if my SL hit, its just 1 losses, not 3 or 4 string of losses like i usually had.
But, ofc not all the high impact news event are tradable, mostly due to spread fluctuation that takes forever to be normal again.

Cheers.

lots of ways to skin a cat ,these are not my charts just taken a interest in learning patterns
you gotta love their anticipation ,while goldman sachs ,bloomberg left no idea

Nice job Bones, valid on weekly also look ECAD.

Dear The Redlion,

interesting topic and your research process seems spot on. I have only 1 question: What sample size(or period of time) would you consider decent in order to evaluate the results as far as is there an edge in technical analysis or there is not?

Best,
H.

https://research.stlouisfed.org/fred2/series/DEXUSEU

that does not answer the question.

not really

Hi Scott there not my charts ,

Interesting topic with some good posts!

It's hard to digest everything but let me put some thoughts in writing. Feel free to discuss them!

There seems to be some agreement that the currency markets can be characterised by 2 distinct phases? A positive feedback mechanism that creates a trend or a random walk with drift. There is also a negative feedback mechanism in which price has a tendency to mean-revert. We could call this a ranging market?

Completely random processes (i.e. a sequence of coin tosses) produces apparant trends and patterns that are in no-way tradable and any profit from them is pure luck?

It has been proven that there is a 'momentum' effect in currency markets, which means that forex is not completely random. Ofcourse the random vs. non-random discussion is somewhat trivial and the profitable non-random market believer has no real interest in the arguments of the other as long as his bottom line is green.

Actual trends in currency markets stem from fundamentals and business cycles which leads technicals. Or they can also stem from psychological factors such as herding? Fundamentals shift and early adopters recognize this. Their behavior creates certain patterns in charts. So, this comes down to what some people seemed to allude to when they claim that over the years you learn to identify certain patterns which is informed money at work. This also reminds me of a another topic on market micro-structure and orderflow (http://papers.ssrn.com/sol3/papers.c...ract_id=966068). Basicaly, this is saying that large institutionals are better informed and the rest of it can be considered 'noise'.

Could we say that the holy grail of trading is really in identifying regime-switches early? I have heard many times that it isn't really that hard to come up with a profitable system for a specific market type. Is the trick not to identify the market type and trade accordingly?

I'm not claiming anything here guys just interested in some discussion and other views.

Best,

positive feedback = trend and negative feedback = range? No!

Let's build a random walk with a very unfair coin. The rules to build the walk are the following:
- The moves are either +1 (up) or -1 (down)
- If Head then the move is the same as the previous one
- If Tail then the move is the opposite of the previous one

This is one possible realization when the probability of Head is 90%, i.e. very positive feedback. Is this looking like a trend to you?

This is one possible realization when the probability of Head is 10%, i.e. very negative feedback. Is this looking like a range to you?

Nice demonstration Pip but your model assumes that the positive feedback or negative feedback relates specifically to the directional signal of the prior bar.

What do the results look like if the positive feedback (relating to a directional bias) remains in force or progressively dilutes over an extended number of bars associated with a single direction that may be reinstated again as another momentum burst? For example where we assume trends may be driven by uni-directional momentum thrusts associated with say periodic news events that underlie the momentum?

More akin to levy flight with non-random positive directional reinforcement over an extended data set supplemented with with random noise on each bar.

For example the chart below is reflective of persistent drift of 2% (over the extended timeframe) with a random element to the price series. It doesn't take into account possible further compounding of drift associated with positive reinforcement over time that would further magnify the directional bias.

Hey PipMeUp,

I only have a basic background in statistics and econometrics but I find this stuff quite interesting! Thank you for your reply.

My statement seems to be wrong indeed. So the positive feedback and negative feedback say something about the persistency ? So, if there is a positive feedback and chances are 90% that price movement persists, isn't this tradable?


How would you statistically define trending and ranging markets? Can we characterize a 'trend' as a drift term in a random walk? What sort of underlying processes are at work in these markets? I'm very interested to read up some more on this stuff so if you would have any good references, i'd be thankful.

Best,

Copernicus you can make the model arbitrarily complex. I just wanted an example of a model contradicting the hypothesis which is both very simply and where the sign of the feedback is trivially obvious.

Per definition it is persistency but it isn't enough to determine if the market is trending or ranging. If the direction changes 5 times over 100 steps (95% chance of keeping the same direction) you can have +20-20+20-20+20 forming a range or you can have +30-10+30-10+20 forming a sequence of higher highs and higher lows. If the change always occured 95% of the time trading would be trivial. But look at a renko chart reversing at each color change doesn't work...

If I give you this sequence of heads and tails: H, H, T, H, T, H, T, H, T, T, H and I ask you what's the probability of the coin gives H? You may think it is 6/11 but it's not that simple. If there is 60% chance the next bar is up there is still 6.4% chance of getting 3 down bars. This sequence can be generated by a coin with probability 70% as well as with a coin with probability 30%. 6/11 is the most likely not the true probability.

On top of that market is more like an unbalanced coin that changes bias all the time. When you have accumulated enough samples to be confident about the probability it has already changed! And I only discussed the direction not the magnitude of the move. You can get up, up, up, big down with a net negative move despite 75% of up's.

You are correct. I am uncertain too. Altough very interesting, practical relevancy is probably limited.

So: Technical Analysis alone will not give you an edge.

I originally started doing some research based on a post in another thread. I found this an interesting statement and went to work in Matlab. My coding skills are not great but I learnt some basics. I hope some of you have some knowledge of Matlab and can correct some of my mistakes. Or perhaps some can try to replicate my results. The results seem a bit too good to be true so there are probably some flaws in there. EDIT: I found a big mistake that changes the results drastically so I'll just post with the corrected results.

Starting from the distribution of price returns in the direction of the trend and against the direction of the trend (The statement in other thread). We need an arbitrary measurement of 'trend' here. Clearly, this is a simplification that I have to make and certainly not very accurate but we had to start somewhere. The other thread uses a multiple time frame approach so I opted for an uptrend if price was above the 10EMA in both the MN, W1 and D1. Thus if price is above the 10EMA, 50EMA and 200EMA on the D1 TF (Another approximation...). Conversely, we assume price is trending down if it is below the 10EMA,50EMA and 200EMA on the D1 TF. (This includes almost 60% of observations so it's pbb too broad if markets really only trend ~30% of the time)

I took historical daily prices from Oanda for 13 currency pairs (Majors and some EUR- and GBP- Crosses) over a period of 15 years (In attachment). I calculated logaritmic returns. Identified when my conditions were met and selected these returns from the full sample of returns. I also generated a series of normally distributed returns N(0,Sigma) with Sigma the sample std dev of all returns (~0.005). This is solely for comparison. I include in attachment 4 series of returns. The full sample, the normally distributed sample, the returns when conditions of trending up are met and the returns when conditions of trending down are met.

We plot histograms + their statistics:


Additonal statistics:

% of observations > 2*Sigma (all - normal - trending up - trending down)

2.48% - 2.3% - 2.9% - 2.1%

% of observations < 2*Sigma (all - normal - trending up - trending down)

2.7% - 2.3% - 2.7% - 2.8%


so I don't see that much differences in % of observations. In a downtrend there seem to be somewhat less % occurences of high positive returns.

What we can see is that 'Trending Up' returns and 'Trending Down' returns differ in their skewness. Mean and Median are also respectively higher then 0 and lower then 0.

I don't really wanna draw many conclusions. I'd prefer to see some validation of these results by somebody else.


EDIT2: In hindsight, there is even bigger mistakes in the following approach and I don't think it comes close to resembling a 'trendfollowing' approach. Feel free to do your own simulations with my trending_returns but don't take the following paragraph too serious... I'll try some more stuff when I have more time


I went a step further then and took my 'trending returns' observations and did a bootstrap. Let's say we go long every day with a new position when the conditions of trending up are met risking 1% and short every day when trending down conditions are met and risking 1% also. I start with a 5000 balance and use 7000 observations for my bootstrap (out of a total 35000 trending observations so this corresponds to about 3 years worth of trading) randomly chosen from the full sample of trending observations. We run the bootstrap 100.000 times. I construct equity curves and save the final equity balance (so on average corresponding to 1 year of trading). The number I get is around 5466 (it's pretty unstable but I think this is the drawback of bootstrap methods?)

Some additional statistics on my trending returns observation:

Wins: 18258
Losses: 16761
Winrate: 52%
Rew/Risk: 1.065

I added my .m matlab code in attachment.


EDIT: So my results are no longer to good to be true. They seem to give an edge but in NO way is this a tradable system.
(Like I said... I don't know if I can trust my results. They seem to good to be true? It would be great if somebody had a look at my code / method or if somebody replicated this. Then, perhaps, we could discuss the implications of this?)

More interesting for discussion is perhaps the return distribution and the skewness? Would it make sense too have more big returns in favor of the trend? Atleast, this is what I would expect?

Good weekend all,

Sorry, I inadvertently deleted my post with the question about the context of the most recent discussion. And thanks tibovwh.

No worries!

Hi There,

My five cents worth " You will never WIN if you are afraid to LOSE ".