How is Statistical Models Programmed and Tested for retail Forex?

Your question is not well stated!

Obviously because I don't know what it is. The term is thrown around here every now and then. Example: "is it based on Statistical Models". After researching this, it could only be one of 2 things. A) System is based on historical testing Odds based systems. Meaning it's got good Maximum Draw-downs, Profit-Factor, Standard Deviation, MAE/MFE, Sterling Ratio, Z-Score, Expectation, ROI, ROR, the list goes on to #-Non-Sense.

 

Or B) they're referring to a type of model like the "Black–Scholes Model". Where you have a Basket of Currencies and looking to capitalize on Mis-Pricing between them. This type of trading strategy is used mostly in Futures, and Advanced Markets and someone trying to explain it here is beyond the scope of this forum. However, I would have at least liked someone to clarify if it's (A) or (B). What comes to your mind when you hear "Statistical Models"?

well here is an example of what a statistical model based trading system is for me. numbers a totaly fiktive so dont try to trade whit them.

you make some statistical investigations and find out that:

if 3 following candles goes in the same direction and the last one's range is smaller that the range of the candles bevor, then in 7 out of 10 times the price will change directions

if price is below 20SMA the price goes int 51% of the time in the directions to the 20SMA

so now you combine this two rules and trade only when they are all true.

.

.in this case you are only trading statisitcal data, no indicator nothing...

Ah, thanks allot zzuegg. Thats sounds like Candle, Elliot Wave or Fibonacci type trading. Only you're creating your own algorithm based on your observation and then Odds. It's all just Filters to my ear. I would only invest time programming and testing such systems if the Time Tested and Popular systems like Candles and Waves do not pan out. This is No secret weapon lol... just the same old same old.

in my opinion it is the same as curve fitting, the assumtion that future markets will behave as same as past one have to be true in order that this system works

By definition, curve fitting IS statistical.

Each day you put the key in the ignition and start the car. One day the car doesn't start so you take the bus. When you get back home and put the key in the ignition do you expect the car to start? Of course not. Past events are indicative of future ones (on average.) On average IS statistical.

Let me take the devils advocate for a second. I'm gonna make the bold statement and say technical analysis is reduced to finding statistical relationships which held true in the past. No one in their right mind is going to trade a system which have text book coordinates for Indicators/Oscillators but fails in back-tests. They're going to optimize it until it yields positive returns. Money and Trade management alone cannot make a non-profitable system profitable. Whats the justification behind any Indicators/Oscillators. IMO Nothing but coincidence on historical data.

 

Anyone advocating against curve-fitting is not far from advocating against Technical Analysis. Fundamental Analysis is not an exact science either. Allot of times News goes one way and the price goes the other. Guess it all comes down to probabilities because no one can define what constitute a curve fit. All we can say is the more filters the more fitted.