Fastest Automatic Trading Software Available for Retail Users

[maxima 10]

There were many comments on "best" software and also on the best for backtesting.

 

What would be the fastest (data feed, calculations and execution altogether) software that is available for the retail traders?

 

Lets exclude latency between execution server and the user computer. Lets just take on account the natural data feed delays, speed of execution and the (least important I think) the speed of the platform itself to calculate the automated logic (mostly it is about the software being able to use compiled code or interpreted code).

 

For instance - Sierra and Ninja use compiled code. What about TS and MultiCharts? Are their languages being precompiled before use and run as a native computer code or they are being interpreted at the time of the execution?

 

What about data speed and execution speed of TS and Multicharts ? What about OEC and NeoTicker? How is Zen-Fire data and execution speeds together?

 

Which combination of data/software/execution broker can be trusted for a system which sends trades every 3 seconds (for example)?

 

I think less than 3s will require a colocation near exchange and the whole lot of other organisation and monetary efforts which I believe will make the project outside the scope of this discussion - DIY automated system?

[Tams 20]

MultiCharts indicators are compiled.

 

My understanding of TradeStation (I could be wrong) is it uses p-code.

 

 

Theoretically compiled code is slightly (ever so slightly) faster than p-code,

but practically, with today's CPU, and with the requirements of trading applications,

it doesn't matter: You won't see a noticeable difference, not even if you are doing multiple regression type of analysis.

 

 

Someone did an experiment on MultiCharts,

he nested 10 MACDs, one on top of the other,

applied it to the same symbol 10 times,

then replicated the set up in 10 charts.

 

ie. the computer has to calculate the MACD 1000 iterations every time a tick comes in.

 

Well.... it hardly budged the CPU.

 

I don't think this is just MultiCharts,

pretty well every charting software can do the same.

If not, blame it on the CPU and buy a more powerful computer.

 

 

 

p.s. one exception might be web based charting software, because they are not designed to do heavy duty work like this.

Edited December 26, 2009 by Tams

[maxima 10]

Tams, do you know any of attempts to compare data feed lags say MultiCharts vs TS vs Zen?

 

Regarding CPU I disagree. 1000 MACDs is nothing in computing time iteslf. Tell the guy to compare 10 Jurik MAs.. He'd see the difference

[Tams 20]

Tams, do you know any of attempts to compare data feed lags say MultiCharts vs TS vs Zen?

 

Regarding CPU I disagree. 1000 MACDs is nothing in computing time iteslf. Tell the guy to compare 10 Jurik MAs.. He'd see the difference

 

 

I don't use Jurik, so I don't know the requirement...

the MACD is just an example done by someone...

Is it representational?

Probably not.

I don't think his original purpose was to make a point,

he was only doing it for experiment's sake,

without any scientific basis or benchmark for comparison.

Does it serve to illustrate a point?

it would only mean something to someone if he think it has value.

Each MACD is 2 moving averages smoothed again by a moving average...

I doubt many real life trader has to drive so much TA and requires so much iteration in one tick.

but then there's always someone doing the unusual thing... so you'd never know.

 

The point is, no matter how slow your software is,

the difference between 1,000 MACD vs 10 Jurik is not material...

it won't be slower than the latency between your eyes and your finger,

nor the latency between your computer and the broker's server.

 

 

I don't know anybody who has tested the Trader/Datafeed/Broker latency.

I doubt any retail trader has the tools, the expertize,

nor all the software/datafeed/broker combination to carry out such a test.

Probably the only people who can do such test is a magazine...

 

 

If you are bend on doing high frequency trading,

the best bet is to do it through a professional trading environment.

e.g. a prop or a fund that has the infrastructure to handle the required speed.

In a home based trading station, or a standard office building,

the Last Mile is your unavoidable weakest link.

 

 

This is only my 2c.

:-)>

Edited December 26, 2009 by Tams

[maxima 10]

no that was the exact point - CPU heavy computations like jurik or some signal processing filters will influent time of the system reaction even on high-end servers, MACD will not even on low-end desktop...

 

and the subsecond delays arent that difficult to measure. if you show me two feeds side-by-side i can tell the difference with 50 ms accuracy.

 

but i have found that multichart can be fed from rithmic, so i have no more questions about it... Would like to know how fast TS feed and execution though...

[maxima 10]

btw tams, i know you are fan of MC. I cant find how MC connects to TT... is it by xtrader API or is it by TT FIX thank you

[Tams 20]

btw tams, i know you are fan of MC. I cant find how MC connects to TT... is it by xtrader API or is it by TT FIX thank you

 

hahah... am I a fan ? ... LOL... I am only a user of MultiCharts.

I have used many other software... they all have strength and weakness... nothing is perfect.

 

Back to your question,

You can connect MultiCharts to TT via a supplied dll.

 

I assume it is through API,

because FIX is very expensive, and it has to be individually setup.

Edited December 29, 2009 by Tams

[Tams 20]

no that was the exact point - CPU heavy computations like jurik or some signal processing filters will influent time of the system reaction even on high-end servers, MACD will not even on low-end desktop...

 

and the subsecond delays arent that difficult to measure. if you show me two feeds side-by-side i can tell the difference with 50 ms accuracy.

 

but i have found that multichart can be fed from rithmic, so i have no more questions about it... Would like to know how fast TS feed and execution though...

 

 

MultiCharts cannot do sub-second analysis yet.

According to one of their forum replies, they are considering adding this feature.

[smuhr 10]

Is Multichart better than Ninja Trader?

 

Tams : whick broker and datafeeder do you use witk MultiCharts?

 

Thx

[Tams 20]

Is Multichart better than Ninja Trader?

 

Tams : whick broker and datafeeder do you use witk MultiCharts?

 

Thx

 

 

MultiCharts and NinjaTrader are TWO very different programs.

 

Both have trial versions, you should download and try them out yourself.

 

 

MultiCharts uses EasyLanguage while Ninja uses C#.

 

There are lots of free EasyLanguage indicators on the web;

C# is more powerful, but also requires more skill to code.

 

Each has its own strength. You really have to find out for yourself.

 

 

p.s. I use InteractiveBrokers for broker and datafeed.

[smuhr 10]

p.s. I use InteractiveBrokers for broker and datafeed.

 

Are you not facing problems for backtesting ?

Because IB has only 24 historical data.

 

You probably import data but from where ?

Thks

[Tams 20]

Are you not facing problems for backtesting ?

Because IB has only 24 historical data.

 

You probably import data but from where ?

Thks

 

 

IB has more than 24 hist data

 

 

besides,

I don't do much long term intraday backtest.

 

The only long term tests I do are with EOD index data, which I can get from Yahoo or Google Finance.

 

e.g.

http://finance.yahoo.com/q/hp?s=^HSI

http://finance.yahoo.com/q/hp?s=^SPX

Edited December 29, 2009 by Tams

[natedredd10 10]

I think its a bit silly to worry about this..it would be to assume your analysis is so precise that adding 1 second to your signal would mess up the back test to any meaningfull degree...You can test this yourself, just add a timer and +1 to your entry signal.

Unless you are trying to incorporate book data it would make far more sense to pick the platform you are most comfortable with as to get the most work done possible.

If you are trying to incorporate book data you will almost certainly need to move up a level beyond retail any, where latency of the feed and execution will become a huge issue.

[maxima 10]

price alone can go 10 tick on NQ in 1 sec several times a day....

 

so I think the question is valid. If not everyone see the value it doesnt mean there is none.

[natedredd10 10]

True, but that is not to say that it isnt random if its going for or against you.

Of course there are strategies that latency would be of the essense, but you should be looking at FIX with a direct connection to the exchange, hosted close to the exchange your trading. It just doesn't make sense IMO to split hairs at the retail level over this at the cost of productivity.

[maxima 10]

Just to prove my point. I have 2.5ms connection to a server which is 0.5ms away from CME and my algo on empty test run can demonstrate 40 microseconds order release. And I still think myself a retail. And the hair splitting I did helped me to achieve this.

 

Sent from my GT-P1000 using Tapatalk

[Chippy 10]

I use Tradestation for all my automated trading and I am happy with them so far. As long as I am making money that is all I care about lol

[TheNegotiator 13]

Just to prove my point. I have 2.5ms connection to a server which is 0.5ms away from CME and my algo on empty test run can demonstrate 40 microseconds order release. And I still think myself a retail. And the hair splitting I did helped me to achieve this.

 

Sent from my GT-P1000 using Tapatalk

 

Where's your algo run from though? If it's London, there's still going to be lag. Fair enough if it's right next to the CME.

Edited May 10, 2011 by TheNegotiator