FXAnalytics - fxan.com

My name is John Putman, some of you that have been around a while may know me and some of my work.

In the coming weeks I'll be migrating a block of research I've been working on since 2009, over to the retail market - this will be the discussion thread. Some of the information will be free, some of it won't. Before we get to far though I need to know I've got something worth selling, that it's of a high quality and is beneficial to the individual trader. The initial stages will hammer all that out and help me tailor it for your use.

This same work is also being readied for institutional applications on a tick level, which brings with it different requirements. If we have money managers that are interested, let me know and I'll get you more detail.

PLEASE NOTE: The website (fxan.com), isn't up and running yet. I want to fully vet the work before I hang a digital shingle out there.

- John

Before we get to far into the background and research I want to start this with my own warning about trading Forex - I'll save the canned disclosures for the actual report.

I've been involved in retail Forex now for ten years and in that time I've watched thousands of traders come and go. Often arriving loudly brimming with confidence only to disappear quietly never to be heard from again.

Making money in this market as a trader is incredibly difficult, takes an enormous amount of time and is inconsistent at best when it is profitable.

Many new traders believe differently and that is often tied into their lack of experience. The more traders you talk to who have traded in real time for five to ten years - who have nothing to gain financially by telling you otherwise, the more you'll hear my sentiment echoed. That is of course if you can actually find them...

You have the right to trade and pursue financial gain in this market if you choose to and it's of interest to you. If I could encourage you though to do anything it would be to do these first:

 

1. Read "The Little Book of Behavioral Investing" by James Montier
2. Read "The Quest for Alpha" by Larry Swedroe
3. Read "Evidence Based Technical Analysis" by David Aronson
4. Once you're certain beyond a shadow of a doubt you have a winning system, trade it religiously and consistently - ONLY - in a demo account for two to three years. Time has a way of unwinding certainty.
5. When you do go live, go small - very small. Mini account small. Don't come out of the gate and try to pay bills, replace a job or do anything that doesn't make very sound financial sense.

If you start there you should be able to mitigate a lot of the damage many traders suffer in the first six months or so of their short trading careers. Making money in any aspect of life is hard, loosing it is easy - Forex is no exception.

For several years I've been staring at a byproduct of my work wondering what to do with it and if it had any value outside of my uses. For one reason or another I never really had time to address it outside of my work - it's been a "back-burner" for a long time.

I work exclusively with computer systems that crunch large data-sets quickly (10 million to 100 million data-points) and generate complex technical trading rules. To get the most out of that data it's important to me to use the most influential parameters possible. So I spend a good deal of time pre-processing the data looking for optimized parameters. If you're familiar with the concept you know this can be fruitless at best and often dangerous at its worse.

The general problem with Parameter Optimization is that it's akin to data-mining and it's ugly cousin - data-snooping, which is only important historically speaking and means little in the future. It's also difficult to do properly: Most single computer setups simply can't handle enough data to prevent the snooping bias or can't do it fast enough to keep the optimized parameters (somewhat) relevant (if at all). (for a good intro into data-snooping and technical trading rules read http://users.telenet.be/webdesignsit...ding_rules.pdf). If you are familiar with Linear Genetic Programming and pumping it through a large GPU cabinet you'll begin to understand the processing power required - at least based on my research.

So what is my experience with Parameter Optimization?
In my area, it works. I can easily measure the impact of using optimized and non-optimized indicators and there is often a distinct improvement using them. This comes with a large "But" though... I've never seen Optimized Parameters - at least in my research - fix a broken system. If your model doesn't work, optimizing the inputs won't help, if it does fix it, it's usually only in back-testing and won't persist as you move to out of sample data or walk forward testing.

My belief, based on my work is, if you start with good parameters and build a solid model from that point going forward you can improve returns and reduce system based volatility. Of course "improved returns" is subjective - a 15% loss is better than a 30% loss (random example).

At the end of the day trading is often about weathering the down-turns to see if the next profitable period will offset and by how much. My bias here is to incorporate every advantage I can for both the positive and negative return periods and see if I can survive to another day/week/month and so on.

The Parameter Optimization Report will be a monthly subscription (more frequent updates when required) that will cover the four major pairs and their optimized parameters for several time-frames and common indicators.

For a while it will be free as stated earlier in the thread I want to make sure it's of value and that means getting it into a lot of peoples hands and then getting their input and opinion. When I do put a price on it, it will be something under $20 a month for the information. I want to keep it reasonable and accessible for the retail clients.

Before I do that though I'll be concluding some work and will post those results for review.

I've spent the week wrapping up some testing and comparison work on the optimized parameters. The good news - in my style of trading they do continue to make a difference. The bad, I don't think they are going to be of any interest to the typical technical trader...

Many traders who employ technical analysis often rely on signals created by a trigger-event (cross over, moving above or below a level, etc.) What has become apparent in the last few days is, many of the best parameter settings would kill a lot of the active systems. As an example, in my model for the EUR/USD, trades taken every fifteen minutes have done the best when using a M/A of 575 periods in the algorithm. An M/A that long resulted in just two to three trades a month when used in MT4 on a M/A cross over system.

The other standout was - and this should have been apparent to me before I started this - when you're trying to avoid curve fitting by using a large data-set, you're going to end up with optimized parameters that work over a very long period. That 575 period M/A worked well over an entire year of 15 minute data. From a practical standpoint it's difficult to justify selling monthly reports when the information never changes!

It was an interesting project, I'm just not sure it's really going to be of any value to anyone.

I thought it would interesting to post a few of the charts from last weeks wrap-up so you can visualize how the optimized indicators impact the trade model. It's not always what you're hoping for, but in my application, it helps overall.

The model here what created on a years worth of 15 minute data and a small basket of indicators. The total size for this test was just over 2 million data-points. That's about a fifth of what I use when I'm creating a real trade model, I'll post one of those equity curves after so you can compare the volatility between this and the other. Right now we're just concerned with the impact of optimized indicator.

Each sample is based on taking a single trade at a single point in time each trading day. As an example the first chart shows the algorithm applied to the the closing price of the fifteen minute time series at 4:00 AM each morning. The actual entry is lagged by fifteen minutes to approximate a realistic amount of timing impact/risk although in reality it wouldn't take that long to execute. The trades in this model were held for approximately four days, this resulted in a 4:1 leverage factor as open positions overlap. In this sample there will be hedges present. In reality the model would simply square itself and I would be neutral until a the model generated a bias. These results then show more spread impact than would actually be present.

This test utilizes M/As for the indicator set. In this data-set P-0 was an un-optimized, randomly generated parameter, which in this case was 18 periods. P-23 was the indicator that had the best fit over the first two-thirds of the data which in this case was 8 months. Again these result are out-of-sample and adjusted for timing impact and spread.

Sample: 4:00 EST Un-optimized

http://www.fxan.com/images/po4AM.png

Sample: 4:00 EST Optimized

http://www.fxan.com/images/p4AM.png

The results in the case aren't typical for my model, this is more improvement than I usually see. Part of that will be do to the amount of time I spent developing the model and the size of the data-set that was utilized.

More to come...

Sample: 6:00 AM EST Un-optimized

http://www.fxan.com/images/po6AM.png

Sample: 6:00 AM EST Optimized

http://www.fxan.com/images/p6AM.png

In this sample we take on a little more volatility in the returns (not a good thing) for about the same gain. This is a good example of no-free-lunch. Nothing works perfectly all the time and everything you do injects its own risk (whether you have accounted for it or not...)

Sample: 8:00 AM EST Un-optimized

http://www.fxan.com/images/po8AM.png

Sample: 8:00 AM EST Optimized

http://www.fxan.com/images/p8AM.png

Now we're beginning to see a pattern appear in the equity curve. Part of this is due to the holding period, which in this model was around four days. That overlap results in most of the samples generating a similar pattern in the E/C. In this sample there is an improvement in the returns.

Sample: 12:00 PM EST Un-optimized

http://www.fxan.com/images/po12PM.png

Sample: 12:00 AM EST Optimized

http://www.fxan.com/images/p12PM.png

Trades taken at this sample actually are better un-optimized. Of course you would never know that until after the fact. We have to look at optimization in its totality to asses (hope) it will work over all.

Sample: 15:00 EST Un-optimized

http://www.fxan.com/images/po15PM.png

Sample: 15:00 EST Optimized

http://www.fxan.com/images/p15PM.png

Returns improved with the optimized indicator at this time period. It's worth noting here again that these returns are being generated by systematically taking a trade at the same time each day and holding in for four days. This of course results in a leveraged position if the trader starts off with a 1:1 initial trade. Based on the volatility of this model it would make sense to start at .25:1 or so to keep it in check. That of course isn't realistic for most of us... The problem is if a trader takes on initial leverage volatility get's out of hand very quickly. That's not just in this model, that's in every model. If a trader started with this at 10:1 by time they hit there fourth trade they would be carrying 40:1 leverage. I don't care how good the model appears to be, that will wipe you out with a quickness on one bad turn.

Case in point - let's say you started trading this model just when it had peaked at around $140,000 (ever buy a fund, a model or a signal service when it looks really good only to have it tank?) You would have had to ride the loss down to nearly $110,000 before your recovery started. If you were trading at 40:1 you would have zero'd your account out before you ever got the chance to recover. Leverage is an account killer, don't let anyone convince you otherwise.

Something else worth noting... This chart shows a leveraged return of 50% for four months of trading. Not gonna happen. When I was managing funds I never, ever took one of these algorithms live and saw the returns I did in testing. There are a lot of reasons for that, some of it even has to do with how you test and so on. But the reality is it just doesn't happen. You have to look at your out of sample results as the best case scenario and hope you achieve just a fraction of that in real-time. If I could achieve a tenth of that in real time I'd be very happy.

Wrong title - this of course in not the 15:00 EST Sample again it's the 18:00 EST Sample

Sample: 18:00 EST Un-optimized

http://www.fxan.com/images/po18PM.png

Sample: 18:00 EST Optimized

http://www.fxan.com/images/p18PM.png

Results improve marginally over un-optimized returns. Overall all in this sample I'd be better off for using the optimized parameter than not, and that is what I typically find for my style of trading.

This a three month out-of-data sample using a large data-set (10,000,000 data points) with optimized indicators. In this sample the trades are held for just a few hours so there is no over-lap and leverage as a result.

Sample: 8:00 EST Optimized

http://www.fxan.com/images/p8amLDB.png

Not quite so much volatility in these returns - but that's never a guarantee.

I wanted to make a general obsevation about the current EUR/USD price action. I've been watching this since early last night and we've done something now that hasn't happened in a while. Typically in a move like this we see volatility stall around the first support band. If the move does continue, it tends to track with the band, which it did for some time. This morning (US time) however, we've seen volatility ramp back up and push through into the second support range.

Historically speaking this condition doesn't tend to persist for an extended period of time. For short term traders this could be a temporary bottom and a good place to start squaring up shorts (next two to three hours). That said, a move into this region can signal a shift in the longer term trend. A pullback could represent an opportunity to re-short as lower-lows are tested.

Volatility levels now are beginning to forcast slowing of price action and consolidation but that could be several hours off.

Very interesting. Though you show charts just of one symbol, I've spotted the word hedge. Is this system based on cointegration testing (johansen/adf) of pairs and curve fitting through linear regression or am I way off. This is something that I've been fascinated with lately and playing with in my spare time.

Hi Ultimate,

I don't use cointegration for my work but I do some equilibrium modeling... ref. Asset Pricing Under Asymmetric Information (Brunnermeier) specifically Bayesian Nash Equilibrium.

Johansen is better suited to what we do but I'm not convinced that profitably trading equilibrium outside of retail spreads is possible. If you're institutional and close to an exchange there might be potential if you can capitalize on a reduction in latency. It's not something I've had time to explore. Let me know how the research goes.

Just a quick update to the current price action. We're seeing things slow way down now. You can never rule out a continuation but I think any short established at this level has a greater risk of consolidation and/or short-term reversal. The next couple of days could be pretty choppy as the current move is evaluated against the fundamentals.

http://www.fxan.com/images/charts/eurusd-4-4-2012-2.png

The pair pulled back into the neutral area of our indicator yesterday and has pushed lower in early trading. I'm not telling you anything you don't already know there...

What's of interest though, are price current levels. They've managed to come down to an area that I would consider "support". Even if we are in the midst of a trend change it will always be with a certain amount of volatility. These levels could signal a slowing of the bearish move today even though it's still early in the US session. That's not an absolute, but day traders should be looking for signs of consolidation here in their own modeling as a choppy bull pullback can't be ruled out.

It's impossible to predict how much the pair could rise if it does pull back, the only thing I would add to that is I'm seeing a base put in on the cyclical indicator and the volatility associated with the bearish move is starting to unwind. This could be an indication that today's or tomorrows reversal will be a little more aggressive than what we saw yesterday.

So... Bearish trend may be emerging - but bullish reversals could have significant volatility.

http://www.fxan.com/images/charts/eurusd-4-5-2012.png

The softening in job growth is more noise than anything else. Any indicator be it technical or fundamental is going to have volatility in it. A good number here or a bad number there is less important than what's going on in the long term average of the release. We'll get to that in a minute...

For the active forex trader the real question right now is, how far is the pull back going to go? Because NFP itself was little more than noise inside a range I wouldn't expect the US Dollar to move more than a little inside of its range. The EUR/USD has quickly pulled back to its 200 period moving average which will also coincide with a basket of other indicators (50% Fib retracement etc). This mean reversion give or take a few pips on either side is likely the majority of the move. Unless of course other fundamental/geo-political data emerges.

The real question now becomes: Do we return to the bearish move put in place early this week and test 1.3000 or will price settle into an area of consolidation? We can't know and your risk here depends on what side of the market you're on and what your typical holding period is. If we start chopping our way above this 200MA and settle into a range there, I would think a test of 1.3000 isn't in the cards for several days maybe more. If we settle in below the 200MA, we could see a choppy return to the bearish trend. Even if we don't, being in a range below that level will make a test of 3000 that much easier.

At the end of the day we have to weigh the current economic condition of the Euro Zone and the US. I think right now in that measurement the unknowns of the EU outweigh the tepid US economy. That may not result in a dramatic sell of the EUR/USD and to be truthful, to this point has only resulted in a lot of volatility on the monthly charts, but it could keep the pair under pressure and any really bullish move subdued for some time.

Volatility and cyclical indicators are heading back into neutral territory, that doesn't mean the move is over but short term traders should be preparing for a reduction in the moves' momentum and calculating their exit strategies.

I'll be back in an hour or so with a look at the longer-term NFP picture.

http://www.fxan.com/images/charts/EU...6-2012-NFP.PNG

I thought it might be interesting to take a quick look at the long term view of NFP. The monthly releases have a lot of noise in them and are significantly less important than what is happening on the "average".

As of right now the payroll data is improving at a decent rate. Of course like any indicator it has a degree of volatility to it that is good for short term headlines, but that's about it. Things are improving, not at a rockets pace, but that's okay too - we don't need an economy overheating into inflation (any more than it is... could I get a break on the gas prices please?)

http://www.fxan.com/images/charts/NFP-4-2012.PNG

The EUR/USD market is taking things a little easy this morning and technical's look pretty neutral. Cross over indicators won't be much help today and any break-out runs the risk of poor follow through as we chop around the 200 EMA looking for a reason to adjust positions.

A good move higher will target 1.3191 while a break lower finds support at 1.2949, both of these are likely too far off though for current conditions.

Current indicators favor range trading over trend, in the short-term anyway. Look for tops and bottoms in your cyclical studies and short holding periods if you're not currently in on a trend move. My cyclical indicator is currently neutral but I would expect to see movement in that in the next 24 hours or so presenting us with something to respond to... - John

http://www.fxan.com/images/charts/EURUSD-4-9-2012.PNG

This is the choppy area of the market I prefer to avoid, it's easy to get beat up here. My volatility measurement shows little potential for a substantial move so I expect things to remain like this through the balance of the day.

What I'm primarily keeping my eye on right now - and it may not be clear for another day or so - is, where do we settle in relationship to the 200MA. If the price relationship to the MA appears supportive I'd start looking for an eventual test of 1.3200. 1.3000 of it settles in below. From current levels that seems like a pretty good gain either way but that's misleading. By time price has actually settled into a range above or below the 200MA half of the move (or more) is eaten up.

About the bear run from a few days ago... It may be coming off the table, that may have been the entire adjustment. At this point the price action doesn't appear to support a continuation of that move and based on my work I think taking out 1.3100 is going to be a lot harder. Time will tell, but we may be in for more noise with little change. That's okay if your a cyclical trader but it will eat your lunch if you're trying to grab a trend. If you trade the "trend" you'll probably want to dial into a shorter time series and adjust your holding periods accordingly.

http://www.fxan.com/images/charts/eurusd-4-10-2012.png