Algorithm/Expert Adviser strategies...

[El Trado]

I'm starting simple on my first Expert Adviser strategy. If the price of a certain currency pair has gone up over the past 3 weeks, then I will buy. If down, I will sell. I will set the SL and TP at equal distances and based on the ATR. This will be a daily candle.

I intend to modify this strategy, hopefully with the educated input of people on this forum. I feel that this very simple logic-based strategy is a good starting point.

Thoughts?

I think it as excellent idea. I did the same thing back in 2014 when I started. My first algo was 200/50 moving average. Obviously not a moneymaker, but it was the first step to where I am today with external investors and a good track record

 

[farmerjohn1324]

After some googling, you can usually find popular aged research papers floating around online: https://pdfs.semanticscholar.org/a204/7655ef23e37c91bfc063a977c558cc9ee903.pdf

As for "EAs that look great in back testing", there are a lot of caveats there:
A) They might not model transaction costs accurately.
B) They might be martingaling (the betting term, you need to check algos that).
C) Working in-sample and out-of-sample are very different things. You can almost always find some periods of past data where a particular algorithm looks awesome, but that don't imply anything about future results.

If authors refuse to give you any rationale for why the EA should work I would stay out of it. In fact, I would stay out of the "EA" ecosystem altogether. What I saw there is mostly gambling via martingaling and grid trading, or using godawful leverages to scale up returns of pitiful Sharpe ratios that easily are noise w/ massive drawdowns, or otherwise implicitly relying on scaling up equity beta (market returns) similarly with the algo just being noise on top.

They do give a general rationale. Sometimes quite specific. They just won't give enough so that I could reproduce their code if I knew how. I never buy anything that uses Martingale. For me, it's 100% automated or nothing due to my lifestyle.

 

[farmerjohn1324]

I think it as excellent idea. I did the same thing back in 2014 when I started. My first algo was 200/50 moving average. Obviously not a moneymaker, but it was the first step to where I am today with external investors and a good track record

External investors?! Wow. Do you need any sort of license for that? I'm gonna guess that you do.

 

[El Trado]

Is there a way to buy/purchase more backtest Forex data? Through MT4, all the brokers I've seen only go back about 5 months on M5 data. I'd like decades if possible.

Download from Oanda thru their API. You can get free data back to 2002 for most forex-pairs

 

[farmerjohn1324]

I don't think there is much juice in forex for the average small retail guy. Currencies are analyzed and arbed up the ying yang by the billions (trillions?) Much easier (but not easy! ) in my little mind to catch some loose dollars with equities. Just my 1c. Good luck.

If I'm going 100% algorithmic, the main difference I see between Forex and stock indices is that the indices always trend up over time, whereas the Forex pairs can be more cyclical.

 

[farmerjohn1324]

Download from Oanda thru their API. You can get free data back to 2002 for most forex-pairs

How convenient that that's my current broker!

 

[farmerjohn1324]

I think it as excellent idea. I did the same thing back in 2014 when I started. My first algo was 200/50 moving average. Obviously not a moneymaker, but it was the first step to where I am today with external investors and a good track record

What was your main source of knowledge for improving your strategy? Was it trial and error all the way? Or did you consult with others for more rapid improvement?

 

[El Trado]

What was your main source of knowledge for improving your strategy? Was it trial and error all the way? Or did you consult with others for more rapid improvement?

Probably 3000 finance books, a masters in data science and a lot of trial and error. (Did not know any data science when I started in 2014). However, I have always been a «finance nerd» and have worked about 5 years in investment banks as a broker as well as analyst.

But I have done the work. To build a successful algorithm is not an «easy way of getting rich» but it is probably more interesting than most other ways.

The reason most people fail is that they want the easy way. So unless you are actually willing to do the work, I would advise you to do something else with your time. A trading algorithm is not a quick way to money

 

[farmerjohn1324]

Probably 3000 finance books, a masters in data science and a lot of trial and error. (Did not know any data science when I started in 2014). However, I have always been a «finance nerd» and have worked about 5 years in investment banks as a broker as well as analyst.

But I have done the work. To build a successful algorithm is not an «easy way of getting rich» but it is probably more interesting than most other ways.

The reason most people fail is that they want the easy way. So unless you are actually willing to do the work, I would advise you to do something else with your time. A trading algorithm is not a quick way to money

Did you decide on just one algorithm? Or do you have multiple?

Do you prefer mean reversion or trend following?

 

[El Trado]

Did you decide on just one algorithm? Or do you have multiple

I have been using a lot. Or maybe more correctly a dozen but with a zillion different versions. Most of the time, I run about 20 versions of my main implementations, to see how the versions are performing. Then I will get rid of the poorest performers and build further on what works