Creating own trading system

[gmst]

Quote from NetTecture:


There is a HUGH productivity difference. NInja was pretty much total standstill. We do more work now in a weekend than we did back then in a month.

Good points NetTecture. Any retail software be it Ninja, Multicharts, Amibroker (possibly) is not suitable to do any kind of heavy-workload like tracking or trading 500 symbols, running multiple optimizations, order book modelling etc. And unfortunately only solution if someone wants to take their trading to next level is to develop a custom built software designed to do exactly this kind of heavy lifting.

This is my opinion however, that such software should be designed only after one has become profitable while using retail level tools. Then trading profits can be put back into development of higher quality infrastructure. All this applies to a retail at home trader. If someone works for an institution, he will have a team of programmers and quants to build pretty much everything that is needed.

 

[NetTecture]

Quote from gmst:

Good points NetTecture. Any retail software be it Ninja, Multicharts, Amibroker (possibly) is not suitable to do any kind of heavy-workload like tracking or trading 500 symbols, running multiple optimizations, order book modelling etc. And unfortunately only solution if someone wants to take their trading to next level is to develop a custom built software designed to do exactly this kind of heavy lifting.

This is my opinion however, that such software should be designed only after one has become profitable while using retail level tools. Then trading profits can be put back into development of higher quality infrastructure. All this applies to a retail at home trader. If someone works for an institution, he will have a team of programmers and quants to build pretty much everything that is needed.

Your opinion is wrong because out of our experience - "becoming profitable" is quite impossible if (a) you never can trust any result you get from a backtest and (b) you simply can not run backtests as you want.

We pretty much spent months fighting ninja instead of developing a strategy. I consider Ninja to be unusable for short timeframe development now (I dont want to say HT - we trade in the minutes range) because it simply is not able to do what it should do.

Trying to finish one optimization for a week fighting with crashes etc. and then having to test every trade manually because oyu never know whether some weeks are just "not loaded" this time is not a way to get productive.

 

[gmst]

Quote from NetTecture:

Your opinion is wrong because out of our experience - "becoming profitable" is quite impossible if (a) you never can trust any result you get from a backtest and (b) you simply can not run backtests as you want.

We pretty much spent months fighting ninja instead of developing a strategy. I consider Ninja to be unusable for short timeframe development now (I dont want to say HT - we trade in the minutes range) because it simply is not able to do what it should do.

Trying to finish one optimization for a week fighting with crashes etc. and then having to test every trade manually because oyu never know whether some weeks are just "not loaded" this time is not a way to get productive.

I have heard many horror stories about Ninja. The issues you mention do not happen with Multicharts. I am a long-term user of MC but as I am becoming a more advanced user, I find myself outgrowing capability of Multicharts. However, for the early days, Multicharts (and before that Tradestation) suited my strategy development needs.

FWIW, I have become profitable using Multicharts developing strategies on minute bars. So, my experience contradicts yours. However, I think it could be simply due to Multicharts more stable and more reliable than Ninja.

 

[NetTecture]

Quote from gmst:

I have heard many horror stories about Ninja. The issues you mention do not happen with Multicharts. I am a long-term user of MC but as I am becoming a more advanced user, I find myself outgrowing capability of Multicharts. However, for the early days, Multicharts (and before that Tradestation) suited my strategy development needs.

FWIW, I have become profitable using Multicharts developing strategies on minute bars. So, my experience contradicts yours. However, I think it could be simply due to Multicharts more stable and more reliable than Ninja.

That is possible - at the time we decided to dump it, Multicharts was not an option (no C# - changed in the meantime).

Point with Ninja is - it was impossible to get any sensible work done. And we had to make one year optimizations - running out o memory 100% of the time for anything else. Work pretty much was on a standstill. Now before holidays we load up the queue and then have the thing run for the days.

That said, grid computing is a must in my eyes these days. Running a 5 year optimization over a dozen computers is great.

 

[Van_der_Voort_4]

Quote from NetTecture:

No. Started working on my own, got distracted, got in touch with some other traders that had started now we have our own inhouse code share solution. The weak point right now is still real trading - they had an approach similar to TradeLink. We started reworking it from the bottom up.... starting with backtesting etc. Now the last part going into rework is trading. So far it is a cmd line application (where you use cmds to start and stop strategies). That has obviously some problems (runs on desktop), so the rework on that will end with a windows service (+ web based control panel). Tons of side features to fix on the way (like isolating strategies into strategy pools that run as separate appdomains).

But no, I do not see anything out there that is decent. It starts with backtesting. Look at tradelink - no optimization is there that is worth talking of, and backtests are not really there either. No cluste,r, not analysis, no data warehouse. Yes, it is open source, yes you can build it all, but that is saying like "hey, I bought a house, now i have to go to a shop, buy the parts and build it". There are some parts, but they are not really "there" in an integrated and usable form.

Hi guys

i really like your discussion

my question :Is iqbroker worth a try ?

http://www.iqbroker.co.uk/platform/trading-software.aspx

it is fairly new platform and there is not much written about it on this forum or any others for that matter

I am studying c# ,in a future i plan on picking a platform and build strategies,

i don't know which one to chose,perhaps iqbroker ,what you guys think?

 

[Nab]

Quote from Van_der_Voort_4:


i don't know which one to chose,perhaps iqbroker ,what you guys think?

At least their webpage is quite amusing:

"IQBroker can be connected to MBTrading and Interactive Brokers for high-frequency trading."

 

[newbie2006]

My 2 cents.....

I tend to agree with hftvol that no ready made software whether commercial or open source will really cut it if we are talking about serious HFT, but yes you can take a look at tradelink to get an idea of what is going on

I've been doing this since 2006 and yes it will take you some time to build your own platform but it will pay off if you have some winning strategies

Another note, when talking about HFT back testing is not a simple task as most people think or suggest and I can give you few examples of why it is not if you want

Anyway I wish you good luck ....

 

[gmst]

Quote from newbie2006:


Another note, when talking about HFT back testing is not a simple task as most people think or suggest and I can give you few examples of why it is not if you want

Please elaborate and give detailed examples. Many people are listening intently.

 

[NetTecture]

Quote from gmst:

Please elaborate and give detailed examples. Many people are listening intently.

Well, you first have to define what HFT is.

For the layman that may be strategies that require a computer in the same city (i.e. single digit latency). Is that HFT?

What about using FPGA's to decode data streams, stock overclocked machines to get an sdge (like from http://www.liquidnitrogenoverclocking.com/)?

HFT in itself is a very tricky thing to define - many people have different ideas what this is.

 

[newbie2006]

Quote from gmst:

Please elaborate and give detailed examples. Many people are listening intently.

well, 2 obvious examples

1) How would you count for your participation in the market, for example you simulate placing 10,000 shares on the best bid, in real time the market will react to this in a way, in simulation it is hard to account for this market reaction

2) you place a bid at specific price then the price goes down and your order is the best bid, how will you simulate the execution ? will you consider your bid executed once the number of shares is taken from the total number of shares at that price ? or will you consider execution when all shares are taken ?