Thread: System from scratch!

Originally Posted by doczero

It took some entries that I wouldn't have- 7/19 , 8/6 , 7/26 and missed a few - but I think that's the human vs automation bit happening.

Hmm Narn, I've been thinking about this - Its hard for me to tell about these entries as I view the chart a lot more zoomed in. I really think your coding is in the right track. Did you run a backtest- did your numbers line up with what I have posted?

For other readers of this forum - I really want to reinforce the purpose of doing this is truly about how to build a system. I really don't want people to read this - latch onto the entry i've posted and attempt to trade it. Without the framework around it, the entry does not make you profitable. I believe its better to latch onto the lessons/method in this thread and use it as a way to test any idea you come up with.

One thing I have is an ideas basket - I put on a post-it note any idea I come up with. When i'm ready to test something - I grab a random post-it note from the pile and start testing.

Are we not considering the "Value" of ADX!! Suppose it is <10 and hooks up to 15 do we enter a trade or wait for it to cross say >40 then only take the trade.
I dont trade Forex so maybe my prmise is wrong, but if I remember the basic of ADX, it indicates a strong trend only when the value is >40

Direction is the Key to Using ADX Correctly

Hey CCMB - have a read of the link above. In the way im using the ADX the value of the ADX is not considered - it is more then angle / or acceleration of the ADX im interested in

Coding the hooks is not so hard. You need to define some rules and test a bit. There will always be situations where a manual/visual decision would be different then the automated system.

Biggest issue I have with ADX (and every other indicator in the book) is that it lags horrible. If you wait for the hook to complete the move is sometimes over. If you narrow down the parameters for a quicker response the hook might not develop.Same with MA's. They lag a lot and generate a late entry.

Using trade management like setting a limit order lower then the price area where the automated entry occurred does work sometimes but more often then not you have a rally that doesn't pullback.

I have tried a semi-automated system with trendlines and patterns but the automated part is so obvious that it doesn't add real value to my system. You still need to analyse everything manually.

I ran the current setup with the automated hook and MA's for a year (365 days back from yesterday) and had a negative outcome. I did use a multiple target system. Will run some more tests later this week/weekend.

The ADX/MA's system is a momentum system but doesn't consider the higher timeframe trend. You might yield better results if you narrow down the trades to with the trend (on a larger timeframe). That's something I do in my own trading method. Start with analysing the higher timeframes (monthly/weekly/daily/240m) and use that information to pick entries on my 'trading' chart. Mostly 60m and 15m.

Yeh - interesting stuff - I have yet to walk forwards out of the year I am testing - so maybe the year 2005 has something to do with it.

Anyway here are the results for the 3 different combo's I tried, I'll focus on one of these (and slightly optimize from there) provided it passes the out-of-sample test on 2006 data.

Results 1: Stoploss 25pips / Take Profit 100 pips



System expectancy: $11.94 (or 0.47R Positive expectancy)

Results 2: Stoploss 40pips / Take Profit 150 pips



System expectancy: $19.13 (or 0.46R Positive expectancy)

Results 3: Stoploss 40 pips / Take Profit 200 pips



System expectancy: $6.36 (or 0.15R positive expectancy)

So with the first step of optimization we took the system from a positive expectancy of: 0.24R to 0.47R

The next step is to check and make sure we haven't curve fit to the in-sample data we have been using (year 2005).

The next step will be to run the 0.47R system (25pip SL / 100 pip TP) through the 2006 data. If it passes through that year with similar results - we can do some slight tweaking of SL /TP then focus on position sizing.

I'll run the 2006 test shortly.

So here are the 2006 results:




And that gives us an expectancy of: $4.23 (or 0.16R)

It would have been very difficult to trade this system through 2006. Throughout the year there were lots of large loosing streaks. going through a lot of the moves were very choppy, and I think with such a close stop whipsaws truely were the enemy.

From here I'll modify stops again. maybe try SL 30 / TP 120.

The data we will be using for in-sample data is now 2006.

One of the conditions of a good system is that the system must not fail during " abnormal" markets such as crashes. In fact, it should make you more money because the markets move fast during such times.

Originally Posted by gannfan

One of the conditions of a good system is that the system must not fail during " abnormal" markets such as crashes. In fact, it should make you more money because the markets move fast during such times.

My view on this is different:

Crashes or glitch days (like we had early may) have little to do with technical analysis but more news or emotion. So I would prefer a circuit braker that prevents the system from trading during these extreme conditions.