Wealth Lab backtesting with Tick Data?

Quote from Volker Knapp:


So WLD3 is able to use intra day data and can be used for intra day trading. Of course I will be answering further questions here if needed.
More...

When I tried using 1 minute resolution data in WealthLab (this was about a year ago, not the current version), loading just a few years worth of data brought the entire computer to a grinding halt and used up all the memory (1GB of ram). In contrast, loading the same data into a custom C program took up exactly 22MB of RAM.

Since I didnt want to test one year at a time, I found that WealthLab was unusable for testing data that has a high resolution. I imagine that Tick Data would be even more memory consuming than the 1 minute bars, and as such I would not recommend using it with WealthLab, unless the last version has had a substantial improvement in the way it allocates memory.

-bbc
 
There is no doubt that a custom C programm is the better option for large amounts of data, also the version 3 has been improved. BTW, I just ran > 17000 days system in four seconds. That is Dow Jones data going back to 1933. Of course a dual system with self tuning takes much longer.
 
Quote from Volker Knapp:

There is no doubt that a custom C programm is the better option for large amounts of data, also the version 3 has been improved. BTW, I just ran > 17000 days system in four seconds. That is Dow Jones data going back to 1933. Of course a dual system with self tuning takes much longer.
More...

Was that on tick data? one minute data? daily data?
 
Quote from cashonly:

Was that on tick data? one minute data? daily data?
More...
I don't know, but I bet it was daily data.

He says:

"... BTW, I just ran > 17000 days..."

nitro
 
Quote from Volker Knapp:

There is no doubt that a custom C programm is the better option for large amounts of data, also the version 3 has been improved. BTW, I just ran > 17000 days system in four seconds. That is Dow Jones data going back to 1933. Of course a dual system with self tuning takes much longer.
More...

Well, that's less than 1 day of ES tick data...

Also, I think it's merely impossible to get intraday data from the 1930s... Well, there's always the 1-900-psychichotline for those thing... :)

:D
 
Quote from Volker Knapp:

There is no doubt that a custom C programm is the better option for large amounts of data, also the version 3 has been improved. BTW, I just ran > 17000 days system in four seconds. That is Dow Jones data going back to 1933. Of course a dual system with self tuning takes much longer.
More...

I don't see how that's particularly relevant to this discussion. Assuming an average of 250 trading days per year, your data consisted of 17,750 bars (71 * 250).

Testing five years worth of 1 minute resolution data, assuming 6.5 trading hours per day yields 487,500 bars (6.5 * 60 *250 *5), which is 27 times as much data as the test that you ran.

What would really help Volker, is if you could load five years worth of 1-minute data into the latest version of WealthLab and tell us how much RAM that consumed on your machine (both actual and virtual RAM). And also if you could run some really basic system (ie MA cross-over) and tell us how long it took to run.

-bbc
 
Reading this thread through for the first time, it occurs to me that there's been some confusion between backtesting intraday data versus running charts on realtime intraday data. I'm assuming that Cash was asking about backtesting.

I've done some work on 1 minute bars in a demo of WealthLab, both ver. 2 and 3, as I was hoping that 3 had changed. The language is very powerful and once mastered seems that it can do almost anything. The problem that I've had is when WLD handles any large file. I'm using files which are five to seven years of 1 minute bars (ASCII data format). WLD takes an eternity to load these files and draw up a chart, even when I specify for it to use only, say 5000 bars. For some reason, it seems to want to take the whole file, index it somehow, then work with whatever number of bars you've specified. I don't think it's got anything to do with indicators, as I've tried this on "blank" charts as well (deleting any code from the "Editor" tab and using that as the baseline).

This was the only disappointment to me for an otherwise excellent program. But it was significant enough for me to stay with my "aging" TS2000i. With that program, I'll routinely backtest and program against some 300,000+ bars in memory. which it handles quite well (perhaps 15 seconds to load the data on a 2.4GHz machine with only 512Meg of RAM).

The only solution that I found to get WLD to work was to create a reduced ASCII historical file. WLD then handled it, but it took too long still.

It's been a while since I've tried AmiBroker. Back then the programming language was very "Metastock" like. You could really open it up by using VB or some JavaScript, but there was no happy medium (I consider EasyLanguage to be somewhere in the medium range).
 
Are the data provided for backtesting also including marketdepth/orderbook changes ?

I'm doing studies on ES intraday with changes in marketdepth, and would be interested in a provider for historical intraday data which includes accurate timestamps and the frequent changes in marketdepth/orderbook.

I guess it would be really a lot of data, and until now I have only used tick-data/IB-historical feeds (recorded by my own application).

If Wealth-Lab or others do provide changing marketdepth/orderbook, how is the data stored ? Binary, CSV or similar ? Is it possible to export data to other formats, or access via API programming ?

Currently I'm making a BeanShell-featured front-end with IB TWS and doing backtesting on limited historical data, but would like other sources, as well for perhaps implementing integrated backtesting with alternative sources from within my front-end program.

The front-end I'm working on is basically a glorified Matrix-window-type (like Tradestation), with a tape-read and some charting integrated.

BeanShell gives all the power of the Java language - including some new upcoming features from Sun Java JDK 1.5 - like (un)boxing etc. Beanshell offers the possibility of running script-files, and interactive scripting through a console (although the provided console is rather limited with regards to multiline statements etc.).

I was planning to do some compression of blocks of data using bzip2 compression, and then being able to run through unlimited amounts of data, aggregating variables and samples while running backtesting strategies.

I guess Wealth-Lab tries to load all the data into memory at one time ? That would certainly be a very heavy operation if they objectify data or something, and don't just treat data as a raw stream (classified though with regards to marketdepth/orderbook data - if that was included). The same would apply if all individual data is made available for immediate lookup/reference internally - i.e if they were trying to get the data structured in some other way than objects.

Nice technical discussions on the Wealth-Lab site, though. To bad the user-gains isn't mandatory/automatic so one could follow the truly good ones, even if they didn't really disclose all of their secrets.

On the other hand, an intraday automatic trading system which is successful beyond average or around 50% of the time is the holy grail for many of us ...
:)
 
You must log in or register to reply here.
Back
Top