5 years and still no strategy | Page 64 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

WRB/MC strategy:Coder needed to semi-automate my strategy 11 replies

strategy builder/tester for new scalping strategy 1 reply

Is opposite of losing strategy a winning strategy? 21 replies

After years in trading, do you still believe in get-rich-fast BS? 98 replies

Fozzy - still using after all these years 2 replies

Attachments: 5 years and still no strategy
Exit Attachments

5 years and still no strategy

  • #1,261
  • May 10, 2017 6:07pm May 10, 2017 6:07pm
Quoting mzvega
Serial Correlation........ The lack of day-to-day serial correlation in auction markets is often the starting point for random market theory. Studies as early as Labys and Granger (1970) confirm the day-to-day randomness. Even in trending markets, the expectation of up price or down price for the next dayis still about 50-50. Even odds in non-trending markets is easy to understand; in trending ones it is not so obvious. A market that is trending up or down does have an enhanced probability in the trend direction; it is just pretty well masked by...

Excellent post MZ. Thanks for contributing. :-)
  • #1,262
  • May 10, 2017 7:57pm May 10, 2017 7:57pm
Quoting VEEFX
{quote} Hi Copernicus, Could you please expand on your understanding of "Law of Large numbers" as it relates to trading. If it is applicable to your journey and discovery, could you also include your own experience (approach, validation, data sets, duration, sampling etc) on backtesting and what you would consider a large enough sampling for adherence to this 'law'. I want to relate your experience with my understanding of how I relate in my trading. To me, law of large numbers relates more to the frequency (i.e. timing) instead of volume/spread/range/sampling...
No probs VEEFX. Big question so will get to it with examples when I finish the daily grind :-)

C
  • #1,263
  • May 10, 2017 9:04pm May 10, 2017 9:04pm
Quoting freakout
{quote} I didn't ask for a ready strategy - I asked for what I can do in my current situation I won't trade with someone's strategy if I don't like it - simple the best results in my trading career I had 2 years ago when I was trading japanese candlestick patterns - I had a time when I was making 15% per day, but I wanted more and more and something wrong happened to my trading...
After reading this thread up to page two and not looking any further... I'm willing to say you are in no different of a spot RIGHT now then you were 5 yrs ago and also when you started this thread. Somebody 'Erebus' offered you a challenge to 'test' yourself and try to help and you totally shot him down. He was even posting in real 'time.' The lack of respect and interest in a post that someone was trying to 'help' you find yourself makes me wonder how the thread has 64 pages.. I hope I'm wrong with my assumption that you are doing better, but I just don't see that happening yet.
  • #1,264
  • May 10, 2017 9:06pm May 10, 2017 9:06pm
Quoting Erebus
Now we are into TUESDAY Frankfurt Open and London is Open, see the top of the chart Is there enough information to setup a trade? What is it? Good Luck {image}
This was one of the posts I was referring too as well.
  • #1,265
  • May 10, 2017 9:09pm May 10, 2017 9:09pm
Rationalist with a correct theory gains knowledge much faster than empiricist. In other words, reading the correct books, correct scientific journals, learning from real mentors is always a better option should it available to you. Trial on error on the other hand takes quite long time for anyone to find a proper method.

Little background info. There are two school of thoughts in terms of how one gains knowledge: empiricism and rationalism; empiricists gain knowledge via experience and observation, a "trial and error" approach, while rationalist think knowing itself is knowable though sheer reasoning and logic, assuming there is any innate knowledge at all. Eg. A baby who never played legos before would throw things around, or even better, eat them for breakfast unless smacked by parents; until she were given instructions by adults on how to build a lego pony there was literally no way for her to achieve that goal, or it would took her 3 years.


Back to trading or anything in life, correct theory of the world is the hardest to come by therefore read and think before you act, wise up and make good judgement before you commit.

As a former empiricist converted rational-empiricist, it took me more than 5 years to learn the above.
No Highs? No Lows? Must be Bose!
  • #1,266
  • May 10, 2017 9:31pm May 10, 2017 9:31pm
Quoting nealpek
If you have the capital to last, you will eventually figure this out, and realize that 17 squiggly lines on your chart are of no use!! Happy trading.....
You do realize that is BLASPHEMY!!
You must be a HERETIC.
Or perhaps you visited an ADVANCED ALIEN CIVILIZATION
My Threads: Trading is as simple as 1-2-3, Highest Open / Lowest Open Trade
  • #1,267
  • Edited May 11, 2017 7:04am May 10, 2017 10:21pm | Edited May 11, 2017 7:04am
1
  • #1,268
  • May 11, 2017 1:57am May 11, 2017 1:57am
Quoting Copernicus
{quote} That's the thing.....where I feel the opportunity lies is in the possible small amount of non-random behaviour that in general is overlooked by a trader seeking the thrills. If there is a small presence of non-random price movement in an otherwise random price series, then like a casino with a small house edge, you can capitalise on this feature by busting through the randomness with a large sample size. For example, if I can find an array of long term return streams with a small edge (say 1-2% each on average over say the last 16 years...
This. But the trouble again lies in the mental ability to ride your system out until the edge becomes apparent. I think that is what kills most retailers. A bad week and they are on to the next system. That is why I advocate trading small enough to allow the edge to play out.

Quoting mzvega
Serial Correlation........ The lack of day-to-day serial correlation in auction markets is often the starting point for random market theory. Studies as early as Labys and Granger (1970) confirm the day-to-day randomness. Even in trending markets, the expectation of up price or down price for the next dayis still about 50-50. Even odds in non-trending markets is easy to understand; in trending ones it is not so obvious. A market that is trending up or down does have an enhanced probability in the trend direction; it is just pretty well masked by...
This is why I follow you MsVega! Somehow I knew this already but you have provided proof. That is a very big deal.
If you are good at something, never do it for free--Joker
  • #1,269
  • May 11, 2017 4:15am May 11, 2017 4:15am
Quoting VEEFX
{quote} I really like this post Mingary. In my 10+ yrs of journey in FX, I have not come across a single such 'bias' that could be exploited repeatedly and consistently. I have over 5 yrs of trading compelling momentum type condition during news which I have had decent success but only about 7-8 weeks in a year and (and 1 week in Feb and1 week in April consistently for 4 yrs in a row). With that said, trying to seek a needle in a jungle of price manipulations is just not possible imo (hence my moderate success with trading events when liquidity...
I am doing the same. I have given up trying to figure out if the price will zig or zag, instead I wrote some code that is 99% position management.
To look for a "winning strategy" that is market based is useless, one that is "account based" is much better given the fact that any price move is both deterministic and probabilistic but cannot be pinned.
  • #1,270
  • May 11, 2017 4:33am May 11, 2017 4:33am
Quoting VEEFX
{quote} Fabulous post Mzvega. ...Wondering if you have done similar drill-down modelling in your research to see if the % increases exponentially via timeframes? For example Monthly (following last month's direction) >> winrate of 50/50 Weekly+current month's direction >>winrate of xx/xx Daily+current week+current month direction: >>winrate of xx/xx H4+current daily_current week+current month direction: >>winrate of xx/xx etc In my research, some pairs have a higher winrate on MN+W1 confluence, some pairs have a higher winrate on MN+W1+D1 etc. If...
Borrowing from what I do know of AMVT, such an analysis might not be possible, AMVT does not consider time frames, the only timeframe employed in its calculations is the 30min TF (should be 25min ideally), in AMVT markets move between a 'balanced' state and a 'trending/unbalanced' state and this can happen in any (varied) length of time. All pairs will exhibit this balanced/trending/unbalanced states, so I doubt any pair offers 'better' win rate (by your definition). Hopefully mzvega will shed more light on the issue.
  • #1,271
  • May 11, 2017 4:49am May 11, 2017 4:49am
Quoting Mingary
{quote} I am doing the same. I have given up trying to figure out if the price will zig or zag, instead I wrote some code that is 99% position management. To look for a "winning strategy" that is market based is useless, one that is "account based" is much better given the fact that any price move is both deterministic and probabilistic but cannot be pinned.
My most resent study, thoughts and observations corroborate your theory of profiting from a random market (for all intents and purposes), I am eagerly anticipating the result of tests I will soon carry out. If test are positive, I will have proven to myself that:

- entries are totally irrelevant to trading, logic says this is as a matter of fact true. Hence one can profit from totally random entries.
- currency pairs and time frames are irrelevant.
- market do not really change.

A long shot I know but thats where the fun is
  • #1,272
  • May 11, 2017 4:53am May 11, 2017 4:53am
Quoting sinqua
{quote} My most resent study, thoughts and observations corroborate your theory of profiting from a random market (for all intents and purposes), I am eagerly anticipating the result of tests I will soon carry out. If test are positive, I will have proven to myself that: - entries are totally irrelevant to trading, logic says this is as a matter of fact true. Hence one can profit from totally random entries. - currency pairs and time frames are irrelevant. - market do not really change. A long shot I know but thats where the fun is
- entries are totally irrelevant to trading, logic says this is as a matter of fact true. Hence one can profit from totally random entries.
==> Yes indeed. What matters is how the trade is managed.

- currency pairs and time frames are irrelevant.
==> only transaction cost matters

- market do not really change.
==> yes indeed
  • #1,273
  • May 11, 2017 6:45am May 11, 2017 6:45am
What if our trading platforms only showed the following ?
Could we still trade it? How?
Attached Image (click to enlarge)
Click to Enlarge Name: 1111.jpg Size: 167 KB
The truth is hidden from you
  • #1,274
  • May 11, 2017 7:27am May 11, 2017 7:27am
For any of you that like to trade the news, the completely bullsh*t unemployment numbers are coming out. Slap out any chart related to usd, take a look at the last couple of hours of trading just before the big reveal, and if price has stalled just before a turn near a bottom or a top. Observe what happens for about 15 minutes after the news, and you'll usually have a profitable trade.
  • #1,275
  • Edited 9:23am May 11, 2017 8:27am | Edited 9:23am
Quoting VEEFX
{quote} Hi Copernicus, Could you please expand on your understanding of "Law of Large numbers" as it relates to trading. If it is applicable to your journey and discovery, could you also include your own experience (approach, validation, data sets, duration, sampling etc) on backtesting and what you would consider a large enough sampling for adherence to this 'law'. I want to relate your experience with my understanding of how I relate in my trading. To me, law of large numbers relates more to the frequency (i.e. timing) instead of volume/spread/range/sampling...
Hi V

I won't drown the thread so if you want to discuss this further....pop over to my thread. Here is a brief idea of what I am talking about. For anyone reading this....please don't assume that this is for everyone. There are many ways to harvest alpha from the market. This is just one of many approaches and will not appeal to many here.

I quite literally apply a rinse and repeat procedure using very simple rules under automation that fit into two broad categories or styles of trade, namely momentum or mean reversion.

The assumptions I make in my processes are as follows:

  1. The rules must be standard and applied to any liquid instrument in any asset class. In meeting this assumption I therefore identify very simple strategies with as few variables as possible that meet the criteria of either momentum (trend) following strategies or counter-trend (mean reverting) strategies. The characteristics of these two broad families ensure that return distributions from each category are uncorrelated with each other. When one family of systems is in draw-down, the other family of systems is in draw-up.
  2. My trading universe used for each of these models is currently 45 instruments across broad asset classes. The instruments themselves are FX pairs and CFD's hence incur holding costs, spread (commissions) and of course a degree of slippage. I have got a lot more work to do in refining my trading universe (better balancing, further diversification into emerging markets and crypto-currencies etc. and ultimately moving off CFD's into futures given the significant trading costs) but at the moment I am investing my efforts in automation and testing the logic of the approach.
  3. I avoid any form of optimisation including walk-forward testing as after lots of effort in this space it led me to great results in back-test land only. Though it is tempting to tweak variables...I avoid any form of curve fitting in this approach. It fits the philosophy of what I am trying to achieve with my 'casino logic' of simply relying on very broad features of the market....namely that serial correlation exists...but I can never anticipate when this is in force.
  4. Each strategy considered for inclusion must as a bare minimum offer positive expectancy according the the testing parameters applied....namely a 17 year or so back-test commencing 1 Jan 2000 to current day.
  5. Each strategy selected must offer very low correlation to the S&P500TR as I also include an allocation of the S&P500TR (namely the Fusex: Spartan 500 ETF) to ramp up returns.
  6. I strive to only select strategies with positive skew however finding mean reverting strategies with this return dispersion is a tough game. I do this for simple risk management reasons so no individual instrument is going to significantly impact the portfolio.


Below is a description of each system currently deployed in the portfolio I am testing.

Attached Image (click to enlarge)
Click to Enlarge Name: Portfolio Composition.JPG Size: 237 KB


.....and the individual system performance metrics and stats.

Attached Image (click to enlarge)
Click to Enlarge Name: System stats.JPG Size: 201 KB


Here is what the return distributions look like of the portfolio.

Attached Image (click to enlarge)
Click to Enlarge Name: Return distributions.JPG Size: 92 KB


If you have a close look at the performance metrics of each system you will quickly realise that in isolation they are horrible performers and nothing to attract any interest from the adrenalin junkie. The reality however is that most traders will not accept that this is the reality.

Below is a scatter plot of CAGR versus drawdown. The blue dots represent each system's contribution to the overall mix.

Attached Image (click to enlarge)
Click to Enlarge Name: Scatterplot 1.JPG Size: 40 KB



But now applying the principles of portfolio magic......this is what you can do with these individual contributions.

Attached Image (click to enlarge)
Click to Enlarge Name: Scatterplot 2.JPG Size: 44 KB


In particular focus on the scaled portfolio option applying allocation principles which offers a CAGR of 13.3% with a drawdown of 49.5%. This is getting close to the performance metrics of the CTA's. I have more work to do to improve the performance metrics but you probably get the picture.

The 32,387 trades (approx 5.2 trades a day) is a sufficient size to me to have the opinion that an edge is in play here. You will note that all I am doing is applying the principle of the Law of Large numbers using very simply strategies that are configured to offer low correlation to the overall portfolio. In essence the Law of Large numbers IMO is simply a catchphrase to state that the closer your sample size increases towards the total population, the more that sample size reflects the features of the entire population. If the entire population is the market itself, then using an approach such as this can capitalize on gross statistical market features and thereby more effectively mimic the behaviour of all participants. The problem in applying a trading logic on a small sample size is that you simply never know whether that return profile was attributed to random or non-random price behaviour.....which also applies to any form of optimisation approach that adopts a predictive bias based on historical data in that it vastly restricts your systems ability to handle market uncertainty.

Attached Image (click to enlarge)
Click to Enlarge Name: Performance 2.JPG Size: 225 KB


Attached Image (click to enlarge)
Click to Enlarge Name: Scaled variable wt.JPG Size: 79 KB


Cheers

C
  • #1,276
  • May 11, 2017 1:52pm May 11, 2017 1:52pm
Quoting VEEFX
{quote} Sorry.. I have no idea what AMVT is
Forgot to state AMVT - Auction Market Value Theory: mzvega has a long detailed thread on it.
  • #1,277
  • May 11, 2017 3:07pm May 11, 2017 3:07pm
Quoting sinqua
{quote} Borrowing from what I do know of AMVT, such an analysis might not be possible, AMVT does not consider time frames, the only timeframe employed in its calculations is the 30min TF (should be 25min ideally), in AMVT markets move between a 'balanced' state and a 'trending/unbalanced' state and this can happen in any (varied) length of time. All pairs will exhibit this balanced/trending/unbalanced states, so I doubt any pair offers 'better' win rate (by your definition). Hopefully mzvega will shed more light on the issue.
This post was supposed to be in response to VEEFX.......however that post has been removed
I apologize for the time it has taken to respond to your post. Sinqua was correct. It's difficult to give a simple explanation, without going into elaborate detail on concepts without providing background. AMVT (Auction Market Value Theory) concepts details a vast amount of research that cant be encompassed in just a few posts. .

First let me start with permitted practices. AMVT does not allow the use of any data that is not "Market Generated info/data". Candles/bars are not Market Generated info..

Market generated data/info is closely tied to something real and observable in the market; whereas an indicator is the result of a computation--typically from a moving average related process, removed from observation.

Using "market generated information" means using only inputs or information provided by the market itself. For example, using only a market's high, low, and closing prices in your analysis.

What makes the high, low, and closing prices market generated info, distinguishable (objective) from other prices? The time in which they occur. Without the element of time, prices would be indistinguishable in all but name. i.e. what makes the closing price distinguishable from any other price on the chart? The time at which it occurred. Every price tick carries two pieces of information, price and the time when it occurred. If your data doesn't include the element of time, it is not market generated data/info

What makes the use of linear price series a "non permitted practice" in Auction Market Value Theory? What makes one price distinguishable from any another price? Again, The element of time. Price over time makes any one price, objective & distinguishable from another. Not by measuring price against a series of other prices.

A moving average related process removes/doesn't include the time element. It's averaging price over a series of prices, the time element is removed. And the process itself comes from "outside the market" (outside/removed from observation, not market generated). Technically being "outside info" (subjective) also a non-permitted practice.

Again, if your data doesn't include the element of time, it is not market generated data/info. A mathematical computation using market generated data" (variables) is a permitted practice

When your computation excludes the element of time as with most technical analysis (i.e. MA's oscillators, etc.), you are essentially measuring(averaging) a "series" of just prices. The only thing that makes those prices indistinguishable from one another is the Time at which it occurred .

Without the element of time, you are measuring "price against price", which has no informational value. Price has no meaning or informational value, unless you know the time in which it occurred.

Measuring one price against another price does not have the same informational value as measuring "price over time""price over time" is not biased by the TF (candle HL) of the observer.

The idea behind using market generated info is to remove your own opinion from the equation and allow you to trade the actual market, rather than trade what you believe the market should be doing. An advantage to using market generated data for your analysis is the results lend themselves well to statistical validation rather than a belief system.........


In my previous post, I demonstrated that using/following a linear series of prices "day to day" skews your perception of the underlying movement in the market. As I explained it only appears to move randomly. You can't use linear measures to describe a non linear environment, doing so as I demonstrated gives you skewed results. All candles/bars are linear measures. Therefore, their use is not a permitted practice in AMVT

The market is not "linear", as I demonstrated in my previous post

Raw market data is not "linear". Market data isn't reported nor does it enter the system in a sequential (linear) series of prices i.e. 1.001 , 1.002, 1.003,x 1.004, 1.005......etc.
Attached Image

Starting with raw data what happens next is the data is "subjectively" broken into TF's ( 1m, 5m, 15m, 30m.etc.) depending on the viewer. This artificially skews the data. Each place where the data has been "subjectively" broken into TF's creates a artificial candle high and a low. This is "subjective" info because it is biased by the observer's choice of TF. Already you have "flaws" in your data set.

Now that the data has subjectively been broken into TF's. The data inside timeframe is now "sorted" (rearranged) into a linear "series" from high to low forming a bar/candle. Now you have skewed your data set once again. Remember the tick data didn't start as a linear series. Your TF's only look like a linear series of highs & lows because it's manipulated the data to do so to create a computer ready format for plotting/charting. This is not "Market Generated info".

Bar chart data, the high & low of each bar is determined by the viewers choice of TF, not set by the market itself, and "price and the time which it actually occurred" is removed by sorting/rearranging the data into a high to low series of prices.

A bar/candles are the result of skewed & manipulated data process. A candlestick "high" is not a "market generated" high. It only "looks" like a high because that's where you chose to break the data into time increments (tf's). The candle open & close are determined again by the viewers choice of TF. The H-L's of each bar are artificially created & determined by the viewers choice of TF, not the market.

I am not saying all TA is wrong, what I am saying is, all analysis that produces results from the use of Skewed data sample is questionable. These are the reasons bar charts & TA are non- permitted practices in AMVT. It's not the methods used in TA that are flawed, it's the data used for analysis, that is flawed. Using TA on timeseries data or bar charts is trading at best with poor data. Following a linear series of prices, skews your perception of the underlying movement of the market as demonstrated in my previous post

But to your question, when using data samples across a correct sample of time (rather than a series of prices candles/bars) as permitted by theory , the informational value starts to slowly erode when you go beyond a 20 day sample rate..

But I see the question is no longer relevant...........
Markets are not efficient, rather they are effective - Jones
  • #1,278
  • May 11, 2017 3:40pm May 11, 2017 3:40pm
Quoting jmn5611
{quote} This. But the trouble again lies in the mental ability to ride your system out until the edge becomes apparent. I think that is what kills most retailers. A bad week and they are on to the next system. That is why I advocate trading small enough to allow the edge to play out. {quote} This is why I follow you MsVega! Somehow I knew this already but you have provided proof. That is a very big deal.
Nonlinear statistics just falls into the same category as chaos theory and quantum mechanics. For a trend to exist there is a subtle component making it unlikely to go in the opposite direction (otherwise it would be a stalemate and the market would never move). This is just an example of what MA's are used for. They identify the trend and how far the current price has deviated(random-walk) from that baseline. In the end; all he did was explain how a Moving Average can be useful and use a precise dataset to demonstrate that nothing in the universe can be quantifiably measured with absolute precision.
"The fun is in the hunt. Not the kill."
  • #1,279
  • May 11, 2017 5:21pm May 11, 2017 5:21pm
Quoting Redeflect
{quote} Nonlinear statistics just falls into the same category as chaos theory and quantum mechanics. For a trend to exist there is a subtle component making it unlikely to go in the opposite direction (otherwise it would be a stalemate and the market would never move). This is just an example of what MA's are used for. They identify the trend and how far the current price has deviated(random-walk) from that baseline. In the end; all he did was explain how a Moving Average can be useful and use a precise dataset to demonstrate that nothing in the...
You are not wrong in what you say. I would add however that MA's do not tell the complete picture. The point being made is that there is no real certainty in the markets. Trends are essentially illusions, they exist because we think they do. This is why markets oscillate. The amplitude and wave peaks and troughs are random. The drivers behind this behavior is not. We have all learned the same TA, FA, and PA. In effect, these things become self fulfilling prophesies. You can flip a coin to take a position, but one must still adhere to the traders equation in that risk must be smaller than reward over a large number of trades. TA and FA give short and longer term direction, PA gives entries and exits. So Ming and others are right in saying while price is random most of the time. Systems are, well....systems. They cannot be random because they are based on a set of rules.
If you are good at something, never do it for free--Joker
  • #1,280
  • May 11, 2017 6:16pm May 11, 2017 6:16pm
Quoting VEEFX
{quote} In regards to law of large numbers, any particular 'strategy' which generates a large number of trades deteriorates performance in my findings. combination of strategies resulting in large number of trades spread across all timeframes and pairs reveals a different story.
I think you hit the nail on the head here.

Let's assume that a system with predefined rules (constraints) is profitable in a defined set of market conditions.

A more prescriptive system (call it system A) with more variables is even more selective in the particular market conditions in which it thrives. It will capitalise on performance during those market conditions (generate strong returns) but suffer during the balance of market conditions (negative returns/drawdown). It's return profile will in general be more volatile. Some great returning periods inter-dispersed with significant drawdown periods. I tend to shy away from these type of systems as I simply can't predict future market conditions.

A less prescriptive system (call it system B) will be able to handle a wider range of market conditions but in doing so will sacrifice the strong returns to accommodate more market uncertainty. It's return profile will be in general less volatile as the positive and negative return periods are diluted. This type of system is my preference as it allows for more robust sustainable performance over a greater variety of market conditions.

By adopting preference B, I can therefore compile a fairly large list of strategies that meet my criteria of positive bias that would be unappealing to participants looking for a holy grail. The problem however is that each system on their own generates weak alpha.

To improve risk weighted returns I simply must diversify...and this is where the Law of Large numbers comes in. What I am referring to here is a large sample of return distributions as opposed to a large number of trades. Each instrument, timeframe, system etc generates their own distinctive return distribution over a defined period. The return distribution (equity curve) is a reflection of how the rules of a trade series (a system) respond to market conditions over a given period.

To ensure that my systems can respond to as much variability in market conditions as possible, I test across as long a timeframe as possible using available market data and specifically look for any positive contributor that is as uncorrelated as possible with the balance of the portfolio.

....so in a nutshell I use the principle of the Law of Large numbers in terms of different classes of return series distributions as opposed to say the number of trades. If you look at the system performance results of the individual systems I posted, this agrees with your logic. Those short term systems with a larger number of trades have worse performance metrics than those longer term systems (that by definition) have a fewer number of trades.

I think we can almost apply a Law here......The greater number of trades for a given system, the worse the performance. The performance of a single system cannot be expected to work all the time. The greater number of trades applied across a range of market condition virtually guarantees worse performance as there is a greater loading of transactional costs applied across a duration of market conditions within which only a small set of particular market conditions respond positively to the applied system.

C
1 trader viewing now