your own trading software
- Thread starter Goldlover
- Start date
NorgateData
Sponsor
For my own trading, developing trading systems:
Backtesting speed (without survivorship bias)
As a tester of systematic approaches since 1999 (albeit with serious naivety then) I'm amazed with the performance of off-the-shelf software can do these days.
Previously, an iteration of "change the trading system to do X" would take several to (several) dozens of minutes to see the result. I remember in about 2002 I did this sort of testing and it would take about 3-4 hours to complete.
These days, I can test a trading system on Russell 3000 constituents (11K stocks) back 30+ years on daily data and it will take less than a minute.
This is a game changer in systematic trading.
Backtesting speed (without survivorship bias)
As a tester of systematic approaches since 1999 (albeit with serious naivety then) I'm amazed with the performance of off-the-shelf software can do these days.
Previously, an iteration of "change the trading system to do X" would take several to (several) dozens of minutes to see the result. I remember in about 2002 I did this sort of testing and it would take about 3-4 hours to complete.
These days, I can test a trading system on Russell 3000 constituents (11K stocks) back 30+ years on daily data and it will take less than a minute.
This is a game changer in systematic trading.
Last edited:
A combination of charting capabilities of Ensign/TradingView, scanning and industry categories of TC2000 (for US stocks), Amibroker’s AFL programming and customization capabilities, plus Trading Box Order Management for scalping.
Priority would need to be no bugs and speed.
Prioritized feature would be unlimited customization as is the case with Amibroker.
Priority would need to be no bugs and speed.
Prioritized feature would be unlimited customization as is the case with Amibroker.
Without vectorization that's about 82 million iterations. At such number the number iterations matter much less than what is actually being done during each iteration. Updating a moving average may add minimal overhead but calculating a correlation coefficient or so can severely add to the time it takes to do that 82 million times. But in the end the computations are the same unless you harness technology that did not exist before (eg, inference a DNN or ML algorithm).
So, what has gotten faster is the hardware, really not software related, at least in most cases. If your algorithm is not path dependent then your CPU will automatically vectorize some computations, making it much faster. But even with path dependencies the clock and bus speed has dramatically improved.
So, what has gotten faster is the hardware, really not software related, at least in most cases. If your algorithm is not path dependent then your CPU will automatically vectorize some computations, making it much faster. But even with path dependencies the clock and bus speed has dramatically improved.
Assuming that is for company:
- Web only. Forget about mobile and desktops apps.
- API, OS and language independent. Rest with Oauth2 authentication.
- Watch-list and obvious order execution buttons.
- I would leave back-testing to a separate realm. It can be an independent app by itself.
- TradingView charts. Just the light-weight library has been giving me a lot of features for free. Many will love indicators. I don't. So a basic chart will work for me.
- Forget about news. News are just a distraction.
- Options chain with a side view of the underlying.
- Ability to publish your trades for inspection. A kind of journal if you wish.
- Gaming-like scores. The best traders get a place in an achievements board. So we can get rid of fake journals.
- Internal user chats. With all the whistles that we are used to in social apps.
- Dividends dates searcher for stocks.
- Order book, ladders.
- Depending on the symbol you trade, official reports summaries as in CPI and Forex calendars. We need to know what the "authorities" are doing to rig the market.
- Retail trader workforce. ability to comment on others' trades. Unite us all.
There is a lot more to add, but these are just a few basic points.
- Web only. Forget about mobile and desktops apps.
- API, OS and language independent. Rest with Oauth2 authentication.
- Watch-list and obvious order execution buttons.
- I would leave back-testing to a separate realm. It can be an independent app by itself.
- TradingView charts. Just the light-weight library has been giving me a lot of features for free. Many will love indicators. I don't. So a basic chart will work for me.
- Forget about news. News are just a distraction.
- Options chain with a side view of the underlying.
- Ability to publish your trades for inspection. A kind of journal if you wish.
- Gaming-like scores. The best traders get a place in an achievements board. So we can get rid of fake journals.
- Internal user chats. With all the whistles that we are used to in social apps.
- Dividends dates searcher for stocks.
- Order book, ladders.
- Depending on the symbol you trade, official reports summaries as in CPI and Forex calendars. We need to know what the "authorities" are doing to rig the market.
- Retail trader workforce. ability to comment on others' trades. Unite us all.
There is a lot more to add, but these are just a few basic points.
Do you know the thing about software like SAP, it's basically fixed and you have to change the business to the software, anyone trying to change the software to the business progressively fails, it's designed to work at $1mil revenue per employee where the more you change it the lower that number becomes.
Now what if trading were no different, there was a fixed set of dynamics and if you try and change the software to the business the probability of failure increases, where the more you adapt the software the more it decreases your revenue per employee, so many what ifs.
Now what if trading were no different, there was a fixed set of dynamics and if you try and change the software to the business the probability of failure increases, where the more you adapt the software the more it decreases your revenue per employee, so many what ifs.
Cross sectional backtesting. For example be able to rank a large universe of stocks with one or more factor rankings using end of day data or even intraday data and then take positions until the next iteration. I have coded up such a backtester in python. You could probably also do it in Amibroker but I imagine the code would be not elegant and clunky.
%%
LOL, no wonder GS,MS, BAC ...... hold massive positions in SAP.
GS average= $997,000 revenue per employee;
BAC average =$448,000
Like NorgateData just noted, include ''1999'' Then 2000+ 2002.........
I make to sure to look@ what IBKR+ SCHW has, including but NOT limited to software.
SCHW has some good stuff like Market Edge; a bit complex , but still helpful.
IBKR also. But peradventure even better than the software;
was the old IBKR magazine paper ad .
The founder of IBKR was on a horse , with a long line of yellow cabs+ bottom line = ''NEVER chose an ordinary investment vehicle'' LOL+ true
Databento
Sponsor
Being able to use the same code for production and backtest is probably the most important.
Last edited: