Probability

I have been look at into the development of a system that is totally indicator/ bar pattern free and fully probability driven, not an easy task, but once created you could in theory place it on to any pair any time frame any broker and achieve better than average results and it would not matter whether the broker alters the price or not so I started with a few “simple questions” what a joke. The more you think about it the bigger the questions become, anyway here are the starting questions
Questions (All based on the position entered on a new bar open)
How do you identify in which direction a bar is likely to move without using indicators?
What is the probability of the TP or SL being hit?
What is the maximum TP will be hit?
What is the maximum SL will be hit?
What can we control?
What is out of our control e.g. volume, price, Etc?

Some of the other things we need to keep in mind.
Should you treat your broker like your friend, or an actual broker or the house as in the casino house because as you all know the house always wins?
What the brokers edge (the house edge)?
The world is run by statistics if you don’t believe me just pick up a North, East, West, South Paper (News paper) and tell me there are no statistics in it.
Why not build a system that caters to that sort of thinking and that it's easier said than done.
Some further reading
http://wizardofodds.com/
http://en.wikipedia.org/wiki/House_edge#House_advantage

Well, the simplest answer is to develop a machine learning algorithm.

Ronald thanks for the reply

Do you have an idea of where we should start.


Here is a somthing I have been working on i figure it's a start. it might be wrong but it's a start

OHLC stands for the 4 elements displayed on a typical price bar:
• opening price;
• highest price;
• lowest price; and
• closing price.

Look at the historical data.
You look at each bar and calculate the how many points are from the Open to the High (Long) and the Open to the Low(Short).
From that I look for the highest probable Take Profit (TP) and the highest probable Stop Loss (SL) for both the potential Long and Short positions
The starting point for the TP and SL calculations is 50 points so the idea is to run a program to find the probability of the price reaching both the TP and SL if the probability is very high I am looking for the system to add a point to the starting point. Then continue to add a point until the probability reaches a predefined point e.g. 0.75
So if I open a position I know I should have a 75% chance of either my TP or SL being hit or whatever I choose
Formula:
n = number of trials this is the number of bars i.e. days trading data
k = number of successes to see if the TP is hit by the Long
n – k = number of failures for the Long hitting the TP
p = probability of success in one trial for the Long
q = 1 – p = probability of failure in one trial for the Long


The formula above should be performed for both the Long and Short side and both the TP and SL

When doing any sort of machine learning, you first need to identify what the machine needs to learn.

1. Are we teaching it to find a specific pattern?
2. Are we trying to get it to predict the future?

Whenever I code EAs, I look primarily for accuracy. Here's what I had in mind when it comes to any form of machine learning.

Instead of trying to predict with any measure of finality whether or not the next bar will close up or down, I'm trying to predict the likelihood of the next bar's closing. If the likelihood is greater than a certain threshold, enter a trade in that direction at the beginning of the bar and close at the end of that bar. Simple as that really. Basic scalping, very low theoretical risk, and should be enough for me to basically throw my entire account at.

With that in consideration, what type of machine learning algorithm should I use?

The easy answer would be some form of neural network. The problem I have with a backpropogation neural network is that they simply take too long to calculate every tick.

Machine learning and artificial intelligence are still very young technologies and have plenty of room to develop. The best place to look for development and advancements in artificial intelligence would be in the gaming industry.

Anyway, I'm still reading up on machine learning and predictive analysis. You're welcome to get started here: http://en.wikipedia.org/wiki/Machine_learning

The most important aspect when it comes to using artificial intelligence for trading is being to process the data as fast as possible.


Regarding your example is a good start, but remember that there needs to be a method for the system to learn from its mistakes. Let the system make "predictions" on the market, and then adjust its score based on whether or not it was right or wrong.

As I see it, the EA would need to be running for awhile if it is to capture enough data (maybe a week) before it can reasonably predict the future. But that's just me.

hey kiwi..... not sure if a direct link would be frowned on, so do a google search for " power stats fx " .....

those "power stats " have been quite handy for me..... they break down probable movement by each pair in each of the daily to 15 minute time frames for the past month, 3 months and 6 months.....

you might have to have an account to see the advanced stats....

they kinda keep my feet on the ground when it comes to profitargets and stoploss's....h

Hah. Even the best indicators are less than 50/50. For large chunks of time the next bar is exactly 50% likely to be either direction.

All a function of volatility and directionality. In a non-directional market it's 50/50 for a 1:1 RR. Which is why some people love the big SLs, since they're less likely to get hit than the TP. Only problem is that provides no edge, one loser trade sets you back multiple winners. Run to a billion trades and you get exactly what you give, minus transaction costs, leaving you with a negative sum game. Hence the importance of an edge... or atleast knowing when an edge exists.

Price / Pip value. I'm not kidding. Price at 1.4? That's 14,000 pips. Can't go any more than that.

The only limiting function is volatility, but volatility has the famous "smile" to it. You get nice easily predictable volatilities, then suddenly... blamo, something unexpected. Volatility regresses to the mean, but the distribution is non-normal.

The distribution actually varies, one day it's a long tail, another it's a normal distribution. If you knew it was a normal dist then you'd fade the extremes, if you knew it was a long tail you'd take the breakout. Sound familiar? But this is not a binary system, it's analog. You've got levels of directionality. You can measure them after the fact w/ a fractal efficiency ratio, or you can design systems to predict it in advance.

When you enter, when you exit, what lot size you use. Everything else is out of your control.

Assuming a fixed statistical distribution, the house can always win. But in the markets, that isn't the case. The house (broker) can most certainly lose. Bucket shops profits regress to the mean too, and without an edge they'll regress to 0 just as we do. The past is littered with failed "brokers."

Their ability to discern trends is their edge. If they can't then they won't offset their risk properly and get hammered.

Yes of course, but not by simple statistics. People try to run simple data crunching to determine the odds of something, not realizing that the extremes don't occur with a normal distribution. There's a 2nd order distribution of odds themselves. "There's a 60% chance that there's a 40% chance that..." kindof thing.

Simple statistical research will help clarify your thoughts, but not provide you a clear edge. The best that can be said is that 2nd order data (volatility, returns, etc) regress to the mean over time. Price data, however, is a different story.

Done and done. Research the market profile.