In today’s article, I want to share with you the results of several tests.
In the last article, I introduced the new features of StrategyQuant 4, which we have been working on for quite some time. One reason is that we have significantly optimized the backtest process itself. Not just parallelization, ie the ability to run multiple cores at once, but also backtest itself.
I have a relatively powerful laptop – Lenovo with Intel I7 processor and 8 gigabytes RAM. I know that not everyone has such laptop, so I did a test on a single core, that means I have reduced my performance to the level of the laptop that most of you have.
Result? Surprising 🙂
Performance of the StrategyQuant 4
I’ve set StrategyQuant 4 to search for the strategies and at the same time proceed with robustness tests. The new version allows you to choose robustness tests you want to perform and evaluation methods directly in the settings, and that makes building process more efficient and saves us a lot of time.
The workflow is following, the strategy only looks for the OHLC of the given timeframe (ie, if the strategy is on H1 timeframe, it tests only for open, high, low and close on H1) and if the strategy is profitable, it tests it also for a more accurate timeframe with M1 accuracy. So the movement on the H1 candle are simulated based on M1.
Accuracy of testing is very important, the use of M1 data on H1 strategy is called inside bar testing and it is key to quality backtest. If you have a backtest without it, it’s more kind of an estimation. Movement simulations based on M1 candles are absolutely crucial.
Let’s go back to our test.
With this setting, we got to the following numbers in one hour and eight minutes:
Number of tested strategies: 61789
Time for one strategy: 66 ms (milliseconds)
Look also at the screenshot where you can see the numbers.
In addition, the required RAM was only 1 GB, which is relatively small.
I’ve also set up very strict robustness tests, so that a single strategy did not pass yet, in this case it is normal. I am looking for high-quality strategies; I do not want to have strategies at all costs that will not meet my expectations. I’ll run the program and report how long I’ve been waiting for the strategy.
Imagine that we can test hundreds of thousands of strategies per day. It gives us a solid lead over those who test everything manually.
The release of the current version is not far. The new version is only available to those who have the full version. If you do not have it yet, we will soon leave the test of the new version to a limited number of testers even without having full lincese of SQ. Follow our site, soon you will find out more details!