Trade based on Autoregressive Model

Thanks MathTrader7 for your nice contribution for EA . i have checked it on daily TF but still it did not put any order from last more than 20 hours .can you guide me

You're welcome! :-)
If you look at the trading algorithm, it won't trade if the time series of close prices (after removing the trend) is not stationary.

Nice work MathTrader ! I wonder though , how come you chose to use AR modelling instead of let's say ARMA ?

You are right! There are more complex stochastic processes than AR(p). However, I chose one of the simplest models to start this thread with.

Hi Mathtrader,

How is this development going?

I'm a bit skeptical about the Dickey-Fuller test for Unit Root. Couple of problems in returns are structural breaks and heteroscedastic residuals.

Could you explain a bit more how your backtest is done?

Best,

Hi tibovwh,

Glad to see that someone is following this track of trading

I examined a set of different models, such as ARIMA(p,d,q). I am busy these day with some other projects. I may publish the results later.
And I see your point about Dickey-Fuller test. I just wanted a unit root test that can be simply implemented into the EA for the backtest. I have recently found out a way to link R language with MQL4, which makes it easier to test other methodologies.

Best,
Math

You can move to a broker like Oanda where you can communicate directly using their API in python. Have you thought about trading in python and using a direct API to communicate with the broker? Metatrader is probably limiting you right now, I bet you could do many amazing things coding in something more direct. In python you have access to tons of libraries that you could use to improve on what you have right now.

Using python language in trading is a very good idea. As you mentioned there are plenty of free Libs for python out there.

Thanks!
Math

I have recently learned about geometric brownian motion. Do you think this is worth pursuing for fitting a model? I think this type of model can come very close to foreign exchange markets and apparently it can be extended with stochastic volatility, heteroscedasticity, fat tails..

What I fail to see right now is how this model will give me an edge? How will I be able to use this in creating a trading system? I'm looking some advice of some more experienced people with a statistics background...

Best,

[/quote]I have recently found out a way to link R language with MQL4, which makes it easier to test other methodologies. Best, Math[/quote]
Hello MathTrader7, would u please tell me about the way you have found for linking R to MQL4?

MathTrader7, you are so smart trader and coder
i wil test your EA now, thank you

Hi Mathtrader7,

I am interested in viewing your EA, I have downloaded it but cannot test it or review it at all?

Hi MathTrader7,

Any chance you could release the code for this?

Hi MathTrader7,

I recently bumped into your thread and it's amazing the great work you are doing here. About that method of integrating R with Mt4, have you made any progress with it? I have been looking for such a method for a long time. I believe that the most accurate and profitable way to trade forex markets is to try to forecast the next day's volatility using something like a jump-diffusion model or a Geometric Brownian Motion. When that has been done successfully, one can then use order flow information to generate buy and sell signals. I'm interested in getting an indicator that can chart the probability distribution of a currency pair for each week. At the beginning of the week, it will use tick price data to calculate and chart the (developing) Probability Density Function (covering from the start of the week, up to the current time period). It will end its calculation at the end of the week and start a new calculation by start of trading during Asian Session on Sunday Evening/Monday Morning. Such a Probability Distribution, ideally should not be a Normal/Gaussian Distribution. It should be one of the Power Law Distributions (e.g. the Pareto Distribution).
I'm in dire need of such Probability Density Function graph and parameters for risk management purposes. I have already achieved astounding results with my methods, but the only issue now is to perfect the risk management using the above-described PDF. I'm not able to code the indicator myself because I'm still learning to programme in C, Python and JavaScript.

Any ideas will be much appreciated.

Metatrader R interface was coded long time ago and modified several times check these links:https://www.mql5.com/en/code/11112
https://www.forexfactory.com/showthr...260422&page=10

This article is trying to estimate the probability density function with expectation maximization algorithm: https://www.mql5.com/en/articles/396
And this is even more interesting article: https://www.mql5.com/en/articles/412

Thanks a lot, olsen-yersen! I will check out those links right away.

I am surprised there are so few ARIMA-based approaches around considering it has been used in financial modelling since at least the 1990s.

These materials are very numerous and obvious. but still I did not understand. What should I do?

Hi MathTrader7, I just stumbled to this old thread and i see that people are mentioning Markov chain/ model which is notoriously hard to find anywhere applied to MT4 , and it worked awesome for Jim Simmons , legendary trader, which made him pretty good returns.


I wanted to share one approach to this model which i found somewhere few months ago on MQL site. Maybe it can be useful to someone or some coder as new trading method or idea for ea. Doing this manually is pain in the ass.

Please look at the attached file for exact details.
In 1940, Wald and Wolfowitz have developed the first and perhaps the simplest market efficiency test - runs test - that is based on the logic of a coin toss series and runs of consecutive heads or tails.

Rules for p-value:
1.Calculate Close minus Open for each period
2.Sign: if Close minus Open is positive, then Sign=1, if it's negative,then Sign=-1
3.Sign change: this is just current period Sign minus prior period Sign
4.Positive runs: Total number of Sign=1 for the selected period. (The period must be user selected)
5.Negative runs: Total number of Sign=-1 for the selected period. (The period must be user selected)
6.Total runs: Total number of both positive and negative runs for the selected period.
7.Positive days: Total number of Close minus Open periods which were greater or equal to zero.
8.Negative days: Total number of Close minus Open periods which were less than zero ie. negative.
9.Total days: Total number of positive and negative days for selected period.
10. Expected runs: Expected number of runs if the market follows random walk for the selected period. Formula: 2*positive days*negative days/total days+1
11.Run variance: Variance of the expected runs f the selected period.
12. Run stdev: Standard deviation of the expected runs of the selected period.
13. Z-stat: Z statistics. of the selected period: Formula: (total runs - expected runs)/standard deviation
14. P-value: Standard normal distribution of the Z statistics. Formula: 1-NORM.S.DIST(ABSOLUTE(Z statistics),1)

Rules for Markov run:
1. Column K: ranking close-open of the selected range.( Range is user specified)
2. Columns L-O: divide C-O into different states
3. Columns R-AP: 5x5 matrix to show how many times different states appear, the matrix is created through extra rows just for an easy excel calculation
4. Columns AQ-AU: sums of different states appearances
5. Column P: sum of rows in columns AQ and AR
6. Column Q: sum of rows in columns AT and AU
7. Column AX Markov run: column Q minus column P

Trading rules:
p-value green: below 11%
p-value red: above 40%
Markov run green: below 0
Markov run red: above 0

Buy rules:
Entry: if both p-value and raw Markov run are green then buy
Exit: if Markov run turns red then exit

Sell rules:
Entry: if both p-value and Markov run are red then sell
Exit: if Markov run turns green then exit

Both buy and sell rules:
1. if entry rules appear several times before exit rules, then those are valid too, but that should also be user specified.
2. All trades should have an individual stop loss, size of stop loss also user specified.
3. Trading times should also be user specified.