Where to find a programmer to automatically backtest a system

[dholliday]

Adding to the above post, the Dukascopy 15 sec data is also missing multiple 15 sec candles (within just one day), so won't be of any use.

All but the most liquid symbols will be missing many minute bars throughout the day, let alone 15-second bars. If you want consistent bars you will need to create them yourself. My platform has a "plug holes" function for historical data. It adds missing bars before the symbol started trading (new symbol), after the symbol stops trading (no longer traded), and missing bars in the middle. You will find that after the 1,000 most liquid symbols, there will be missing days in a few symbols. This is not a mistake, they had no trades. Of course, during live trading, I build missing bars on the fly from the tick data and a 1sec message.
The commercial trading platforms are toys and writing your own platform is a major job even for a very experienced programmer. If you can find something that does most of what you want, and you can program the rest, go for it. Howard Bandy had some books that he shared his AmiBroker and python code. Something like that may get you to where you want to be.

 

[Astrevea]

Excel may not work for you.
If using intraday data, that will be a lot of data going onto a sheet if you are thinking of accumulating multiple days.
Ok, if its just some columns of raw data, but as soon as you decide to add formulas to number crunch your data, the sheet will probably freeze due to overload.

It doesn't freeze but it is very slow indeed. I am still undecided about what I am going to do about all this.. the reasons are in the rest of this post.

I'm always interested in doing projects like this but not for 2-300 dollars. It takes time to do things properly and optimise performance. Not to mention I would need to borrow from my own code the functions needed for many of the elements in the strategy.

Where do you live and would you be willing to work via upwork etc?

If you can't find someone to do this in a few months you can contact me, I may have time. I might do it for free but I could only give you results, not the code. My trading platform is meant for trading, not backtesting. I have written backtesting in it but the process needs to be simplified.
I use DTN/IQFeed so 1min testing data is no problem.

Thanks for letting me know, could you tell me why you are willing to do it for free? I am assuming you are already doing well in the markets, i.e. consistently profitable, etc.. Is it to find out about alternative methods of trading or you are so good at programming that it will take you no time?

All but the most liquid symbols will be missing many minute bars throughout the day, let alone 15-second bars. If you want consistent bars you will need to create them yourself. My platform has a "plug holes" function for historical data. It adds missing bars before the symbol started trading (new symbol), after the symbol stops trading (no longer traded), and missing bars in the middle. You will find that after the 1,000 most liquid symbols, there will be missing days in a few symbols. This is not a mistake, they had no trades. Of course, during live trading, I build missing bars on the fly from the tick data and a 1sec message.
The commercial trading platforms are toys and writing your own platform is a major job even for a very experienced programmer. If you can find something that does most of what you want, and you can program the rest, go for it. Howard Bandy had some books that he shared his AmiBroker and python code. Something like that may get you to where you want to be.

The data source I was trying to use has "filter flats" option, if it is unchecked it is supposed to give you all the candles, including the ones where no trading occurred, I am thinking they are just not a perfect data source, maybe that is the reason.
I researched the Interactive Brokers option, and you can get historical data from them, but for candles of 30sec and less you can only get the history for the last 6 months. Above that you can go back many years.
The method I want to backtest is all about "aggregating" M20-M30 (and all TFs in between with 15 sec increments) candlesticks from 15 sec candlesticks, to capture the maximum number of trading opportunities. Ideally even 5 sec candles. By aggregating using 1 min candles there will be a huge number of opportunities missed (the majority of them).
I am really surprised you have your own trading platform. What broker do you use by the way and what markets do you trade?

 

[Astrevea]

Also I wanted to write a separate post about this candle aggregation/creation concept that I keep referring to, I have no idea what percentage of traders on this forum use this concept, maybe a lot, maybe very few. I am new here... I am sure you can create a set of rules without the candle aggregation technique and trade that way successfully, but I have thought long and hard about it, and for me this is what makes the most sense. Once the M20-M30 candles are aggregated/created from smaller TF candles, I am looking for a 7 candle formation, there are rules for each of the 7 candles (purely in terms of its relative HtoL and what percentage of the candles HtoL is its body). One candle should be a shallow pullback (38.2% at most, but a round number such as 40% can be used, doesn't matter much). Also the first candle of the 7 candle formation should be large (I use the previous 15 historical candles of the same TF and this candle should be at least twice the size of the average of those 15 candles). There are some more rules, but I have already specified some of the key ones. Also I know this method works, but the whole point is to find out just how well it works.
At the moment it seems like I am going to have to use IBs 15 sec/5 sec historical data, which is only available for last 6 months like I said, therefore the number of samples won't be very large, requiring me to go live with small risks initially, and as I continue forward testing it, there will be more and more stats on the method.

 

[dholliday]

I am really surprised you have your own trading platform. What broker do you use by the way and what markets do you trade?

Interactive Brokers with IQFeed data. Stocks only.

 

[dholliday]

Thanks for letting me know, could you tell me why you are willing to do it for free? I am assuming you are already doing well in the markets, i.e. consistently profitable, etc.. Is it to find out about alternative methods of trading or you are so good at programming that it will take you no time?

I actually have several trading systems that aggregate ticks into one-minute bars and one-minute bars into various length bars (though I haven't found anything less than 3 min useful), starting at different minute intervals.
Coding on this platform is quite complicated. Over time I have made it easier. and will eventually have some templates for different types of systems. Backtesting requires too much custom code. There is a bunch of stuff needed to make it easier. I don't really need more work to do, but some of the stuff I will be testing for myself is similar to what you're doing anyway. Currently, I don't plan to use sub-one-minute data for testing so maybe it's not a good idea anyway.
Some day I may want to switch to NxCore data, then I could build everything from the raw tick data and testing would just run the same code. Oh well, for now, DTN works great for me.
Good luck.

 

[dholliday]

Also I wanted to write a separate post about this candle aggregation/creation concept that I keep referring to, I have no idea what percentage of traders on this forum use this concept, maybe a lot, maybe very few. I am new here... I am sure you can create a set of rules without the candle aggregation technique and trade that way successfully, but I have thought long and hard about it, and for me this is what makes the most sense. Once the M20-M30 candles are aggregated/created from smaller TF candles, I am looking for a 7 candle formation, there are rules for each of the 7 candles (purely in terms of its relative HtoL and what percentage of the candles HtoL is its body). One candle should be a shallow pullback (38.2% at most, but a round number such as 40% can be used, doesn't matter much). Also the first candle of the 7 candle formation should be large (I use the previous 15 historical candles of the same TF and this candle should be at least twice the size of the average of those 15 candles). There are some more rules, but I have already specified some of the key ones. Also I know this method works, but the whole point is to find out just how well it works.
At the moment it seems like I am going to have to use IBs 15 sec/5 sec historical data, which is only available for last 6 months like I said, therefore the number of samples won't be very large, requiring me to go live with small risks initially, and as I continue forward testing it, there will be more and more stats on the method.

Since you know it works and you use IB, their paper-trading account is very good. Forward test. You will see the slippage and commissions. If it works, great. This is what I do.
Best of luck.

 

[fan27]

Hi guys,
I will try to keep this as concise as possible. I have a trading method which is as simple as "if price is moving confidently we take a position and when price is no longer moving confidently we close a position", there is a strict set of rules for "moving confidently". This is not the point of the topic so won't go into much detail.
I backtested it previously in a discretionary way (with slightly less strict rules) and it works.
I need to do an automated backtest of it, which I have already tried using Excel, but I quickly realized I won't be able to implement it efficiently due to lack of knowledge of things other than basic IFs, ANDs and ORs... Furthermore, Excel is very slow, even with what I have managed to implement so far.
I use a candle aggregation method to create more trading opportunities (where we use smaller TF candles to construct a variety of larger TF candles with different opening times, i.e. M1 candles to produce M5 candles opening at 9:00, 9:01, 9:02 etc) and so there is a lot of data as well.
Does anyone know where I can find a decent programmer who has background in trading as well? I have already had bad experience with a website called Fiver, got a full refund that time. I saw some people recommended upwork.com, but others were saying it is pretty rubbish for this kind of things. Any further advice on this?
Many thanks.

What is your budget for this project?

 

[trader221]

Learn coding. Quantconnect is a best way for learning ,coding,backtesting and runing algorithmic trading .
I am programmer and i Can help so i speak not english vert good

 

[Astrevea]

What is your budget for this project?

It is difficult to say right now, I am willing to pay a reasonable amount, but what a reasonable amount is I cannot say at the moment, I have been busy with other things and haven't yet tried upwork (which I know some people are not kin on). I will create a request there today and see if anyone replies.

Learn coding. Quantconnect is a best way for learning ,coding,backtesting and runing algorithmic trading .
I am programmer and i Can help so i speak not english vert good

Thanks for your advice, I understand it is beneficial, but I have too many other responsibilities in life now to be able to dedicate myself, especially to such a level where I can run a robust backtest and then create a robust trading algo.

 

[fullautotrading]

It is difficult to say right now, I am willing to pay a reasonable amount, but what a reasonable amount is I cannot say at the moment, I have been busy with other things and haven't yet tried Upwork (which I know some people are not kin on). I will create a request there today and see if anyone replies.

Thanks for your advice, I understand it is beneficial, but I have too many other responsibilities in life now to be able to dedicate myself, especially to such a level where I can run a robust backtest and then create a robust trading algo.

Backtesting is pretty useless and even detrimental. One needs a throughout simulation study to have useful insight into the avg behaviour of your approach.