Hypothetical Research Journal

I’ve been trying to get back to this project for a long time; I think I’m finally seeing a window of opportunity to pick back up in my model research and analysis. This will be a slow grind but I’m looking forward to it…

The first thing I need to determine is the overall goal of the model if it should be launched into real time. While there is very little chance I’ll ever return to professional trading, building a trade model simply for myself, makes very little sense. My risk profile has changed dramatically in the last few years and any allocation to alternative investments would have to be very small. The time invested in developing the model and managing it, would not likely be justified by the returns associated with it; at least not for years and years to come. So the model will be shaped as if it is going to be used for a group of investors.

If the model is going to “hypothetically” be used for investors, it will need to meet certain criteria.

First it should have a low correlation to other models currently in use.
It’s risk profile should allow it to generate reasonable risk adjusted returns.
The model should be straight forward and relatively simple to execute.

FAIL!
FAIL!
FAIL!

No matter how many different colors you pull out of the box a crayon is still a crayon and at the end of the day almost every model can be deconstructed as a trend model; the only thing that really varies is the time frame in which it’s traded.

Reasonable risk profiles are rarely reasonable in hindsight and anyone who is interested in such an approach tends to be broke as they’ve already lost their money on the “less than reasonable” risk profiles.

If trading were straight forward and relatively simple 95% of all traders would succeed instead of fail.

Now that we’ve established some solid baselines I can move forward with the process. The first thing I have to do is get my data-set organized and decide where to put my efforts. Having already been through a lot of this before I’ve got some ideas but want to consider the alternatives as well as so there will be some new testing as we move forward. I’ve also grown as a trader and have insight that I didn’t previously; I want to make sure that my model has matured with me.

Coming up… Trade data, time frames and whatever else makes me happy.

Over the years I’ve been asked a lot, “What time frame should I trade” or “What time frame is the most profitable”. There is a mixed answer for this but it applies to my data work so I’ll take a stab at it. Typically the longer the better… The primary reason for this is there is less transactional impact on your returns and losses. Anytime you can minimize transactional, timing and market risk, it tends to have a net positive affect on your trading. Sometimes it’s the difference between blowing your account up in three weeks as opposed to two, but you get the point.

The problem with longer time frames is that they subject us to higher and higher levels of volatility, which results in deep draw-downs when we hit the inevitable losing streak. Some of that is the result of the leverage typically employed by the retail trader which magnifies the problem. So the answer to the question tends to vary and it has a lot to do with your account size and the amount of leverage and margin you are required to use to make the minimum trade…

Maybe the question should be, is the time frame I’m using working now? Think about that for a second; instead of hunting for the end all be all time series on a chart (which doesn’t exist) you choose a time series that you can stomach the volatility on and then find ways to measure its efficiency for the current volatility profile in the market. Make sense?

Here’s an example: Let’s say I like trading off a two hour time series. It's the longest time series I can trade and still survive the volatility swings. What I need is a way to help identify if that time series is currently working right now, so I know if I should adjust my exposure or possibly limit (or even halt) my trading activity.

Several years ago I began working on this type of optimization for the model and came up with some interesting indicators. I never got the chance to test them in real time but will take a look at them as the model and approach are updated and tested here in 09.

There is also another obvious approach here; if we can with any certainty (hahahahaha…) determine what time series is the most efficient right now, maybe we simply adjust our approach to that horizon… That almost seems to easy and there is no doubt some type of lag in that decision making process, so we may just find ourselves moving from one dead zone to another, but the idea is interesting and may warrant additional thought.

Going back to the original idea; look for a time frame that fits you, your trading and your lifestyle and then look for ways to determine when you can trade it and when you should stand aside.

I organized the data-set today and grabbed a random time-frame to establish a baseline. I used a watered down version of the trade model, utilizing the exact same entry and exit rules over a ten year period.

The idea is to test different parameters within this time frame and then in time series above and below... The initial results were problematic; the returns are too good... There simply is not enough volatility present over this period of back testing. As a rule if a model returns 15% you can expect to see at least 8 to 10% drawdowns, if not higher, through any given trading year... that's not present here, anywhere. Any time you see returns that are substantially better than what is being generated in the "real world" you have to be extremely cautious about what you are seeing.

The first thing I'm going to do here is scrub parts of the data against another set of data to be certain is good. I don't think that's it but we'll start there. The next thing we'll explore is the code that generates the returns, to make sure it's doing it's job properly. Eseentially I'll double check everything and try to pinpoint exactly where the excessive returns are coming from and what has been missed.

Sometimes it takes walk forward (demo) testing to really pinpoint; things change in real time and flaws in an approach can be come very apparent after just a few weeks of grinding out phony returns...

http://www.fxan.com/images/index.12.gif
http://www.fxan.com/images/index.13.gif
http://www.fxan.com/images/index.14.gif
http://www.fxan.com/images/index.15.gif
http://www.fxan.com/images/index.16.gif

http://www.fxan.com/images/index.17.gif
http://www.fxan.com/images/index.18.gif
http://www.fxan.com/images/index.19.gif
http://www.fxan.com/images/index.20.gif
http://www.fxan.com/images/index.21.gif

http://www.fxan.com/images/index.22.gif