Holy Grail Trading Software

In reply to some of the early posts: The only real reason to develop you own trading software is to eliminate arbitrary limitations set by software vendors such as Equis and TradeStation. These days there is little money to be made being a software vendor ... the real money is being made with quote services or brokerages.

I used to be an avid MetaStock user and then by chance went to a TradeStation seminar. It blew me away how many strategies I could implement in TradeStation and not in MetaStock. At that point I decided to simply eliminate 3rd party software and just write the trading software myself. It's probably taken 3 man years (or more) and several versions, but today I have exactly what I want.

I don't think it ultimately makes much difference which programming language one chooses. I started in C++ and have since then moved to Java. The ability to express the problem space efficiently is orders of magnitude more important than saving a few processor cycles here and there.
 
Metooxx,

Sorry, there are really two questions I had:

What is the proportion of computation spent analyzing real-time data versus past data (backtesting, etc.)? Now, considering only real-time processing, what types and quantities of CPUs are you using, or how many numerical operations per second are required? Is your code lean or does it perform calculations with excess precision?

About short-term versus long-term:

I was asking about being dependent on execution-speed. Maybe I wasn't very clear. I was not questioning the value of a fast data feed and rapid trade execution. Those are always important. I was asking if execution speed is a _necessary_ requirement for your expected profitability and minimal risk. Your systems produce trading signals or forecasts. What is the useful lifetime of that information? Minutes? Hours?

My research suggests that shorter term market behavior changes the most over time. This is easy to test. I can take an indicator and evaluate its ability to forecast various future quantities. Since I'm looking across 5K stocks, I can ask: "what is the correlation coefficient on a certain date for this indicator providing an N-day forecast into the future?" I observe with shorter-term forecasts that the average correlation is higher, but the variation of those correlations day-to-day over several years is also quite high. I.e. the market may reverse its dynamics. However, with longer-term forecasts, the average correlations are low, but the variation is also low. What does this mean? A low variation tells me that the correlations are _stable_ over time (low risk).

This goes beyond correlations. I see the same effect in out-of-set trading risk, in terms of STD of equity changes and drawdowns. It's just easier to measure with correlations than with equity-change statistics. Run a test like I've described. It should be obvious. Low correlations may be un-tradable. But, that's either an amplification, diversification, or profitability issue, something that may necessitate a creative trading strategy or the use of derivatives or hedges. It is not a stability (drawdown risk) issue that kills your system.

You often read anecdotal experiences from short term traders observing that "markets are changing", such as how decimalization made Nasdaq scalping more difficult. A trader needs to constantly adapt. Of course! Yet what is happening with long term speculation? You can still pick apart a company's books, try to understand economic trends, and separate facts from BS. The Warren Buffett strategy will always work to some degree. I'm not saying it's easy. But if you do it well, you can trade longer-term positions profitably and not have to fundamentally alter your method of analysis every few months. Yes, everything changes. But, economic and fundamental analysis isn't changing nearly as fast as some short-term strategies. Financial markets are always going to obey common-sense economic and fundamental factors. Quantitative or technical analysis can always pick up on these factors, regardless of the time frame. Do you agree?

Jay
 
This was not clear:

"It is not a stability (drawdown risk) issue that kills your system"

I should have more simply said:

Low relative variation of correlations over time seems to dictate low drawdown risk, even if the correlations are weak on average. Drawdown risk is what kills systems.

Jay
 
Originally posted by nitro


Ritchie ?

nitro
More...

No, maybe the first part picking up "nickels and dimes" was maybe Myron Scholes in a televised interview about LTCM, the second part " in front of a steam roller", I heard from a guy who heads a clearing firm; he attributed it to a guy running a private arb shop
 
Originally posted by jaypaul
Metooxx,
What is the proportion of computation spent analyzing real-time data versus past data (backtesting, etc.)? Now, considering only real-time processing, what types and quantities of CPUs are you using, or how many numerical operations per second are required? Is your code lean or does it perform calculations with excess precision?

I was asking about being dependent on execution-speed. Maybe I wasn't very clear. I was not questioning the value of a fast data feed and rapid trade execution. Those are always important. I was asking if execution speed is a _necessary_ requirement for your expected profitability and minimal risk. Your systems produce trading signals or forecasts. What is the useful lifetime of that information? Minutes? Hours?

My research suggests that shorter term market behavior changes the most over time. This is easy to test. I can take an indicator and evaluate its ability to forecast various future quantities. Since I'm looking across 5K stocks, I can ask: "what is the correlation coefficient on a certain date for this indicator providing an N-day forecast into the future?" I observe with shorter-term forecasts that the average correlation is higher, but the variation of those correlations day-to-day over several years is also quite high. I.e. the market may reverse its dynamics. However, with longer-term forecasts, the average correlations are low, but the variation is also low. What does this mean? A low variation tells me that the correlations are _stable_ over time (low risk).

You often read anecdotal experiences from short term traders observing that "markets are changing", such as how decimalization made Nasdaq scalping more difficult. A trader needs to constantly adapt. Of course! Yet what is happening with long term speculation? You can still pick apart a company's books, try to understand economic trends, and separate facts from BS. The Warren Buffett strategy will always work to some degree. I'm not saying it's easy. But if you do it well, you can trade longer-term positions profitably and not have to fundamentally alter your method of analysis every few months. Yes, everything changes. But, economic and fundamental analysis isn't changing nearly as fast as some short-term strategies. Financial markets are always going to obey common-sense economic and fundamental factors. Quantitative or technical analysis can always pick up on these factors, regardless of the time frame. Do you agree?

Jay
More...

1) Virtually all is real time, we test against a live feed not historical, then we test with real money.

2) Not a PC, the answer is proprietary.

3) Lean.

4) Execution speed is paramount, sub second to a few seconds, rarely minutes.

5) Agree shorter term signals come and go, but when you find them they are more reliable; until they quit working.

6) I am sure long term speculation works, however it is not our game, I would rather make XXX% a month on a fraction of the capital that would be required to make 25% a year.

We come from a different game than you are talking about, we are arbitrage traders, the profit is on the screen before you do the entry order.
 
Originally posted by vikana
In reply to some of the early posts: The only real reason to develop you own trading software is to eliminate arbitrary limitations set by software vendors such as Equis and TradeStation. These days there is little money to be made being a software vendor ... the real money is being made with quote services or brokerages.

I used to be an avid MetaStock user and then by chance went to a TradeStation seminar. It blew me away how many strategies I could implement in TradeStation and not in MetaStock. At that point I decided to simply eliminate 3rd party software and just write the trading software myself. It's probably taken 3 man years (or more) and several versions, but today I have exactly what I want.

I don't think it ultimately makes much difference which programming language one chooses. I started in C++ and have since then moved to Java. The ability to express the problem space efficiently is orders of magnitude more important than saving a few processor cycles here and there.
More...

this is what eventually happens; either one can find a package that takes them most of the way, and suppliment it with another package that completes their "gap" in features as such, or one can take a baseline service (Market Data Vendors/Feeds) that allows custom development and then add in both a Mathlab or C+ routines for calcs and a charting package (specializing in GUI/graphical displays) which is highly customizable.

in the same time that it takes to adjust your trading chair to the desktop, the markets have recovered, attained new highs and those just using simple tape reading and free internet financial websites would have surpassed them.

the challenge has always been the "love of the game" as evidenced by programmers, software designers and such, of which it appears most of us are, at least those brave enough to expose their opinions; and the challenge of "love of profits through trading" which focuses one's attention on making the wave count instead of how one discovered the wave, and how one executed one's orders in time with the wave.

there's no happy medium.

Cheers to the starter of this thread for being brave enough to expose his efforts to the light of day and to be respected by the contributors of these comments. That was no small feat....
 
You must log in or register to reply here.
Back
Top