Comments on: Why 90% of Backtests Fail https://financial-hacker.com/why-90-of-backtests-fail/ A new view on algorithmic trading Fri, 03 Jan 2025 12:59:29 +0000 hourly 1 By: jcl https://financial-hacker.com/why-90-of-backtests-fail/#comment-113981 Fri, 03 Jan 2025 12:59:29 +0000 https://financial-hacker.com/?p=4373#comment-113981 You can do both, but normally it’s for all components.

]]> By: AT https://financial-hacker.com/why-90-of-backtests-fail/#comment-113980 Fri, 03 Jan 2025 12:37:09 +0000 https://financial-hacker.com/?p=4373#comment-113980 In a portfolio system, do I have to do the MRC for each component separately or can I do the MRC with all de components of the portfolio at the same time? Thank you.

]]> By: Trend Following Done Right https://financial-hacker.com/why-90-of-backtests-fail/#comment-112248 Sat, 07 Dec 2024 14:09:24 +0000 https://financial-hacker.com/?p=4373#comment-112248 […] are also “reality checks” that reduce the chance of data overfitting. Here is a way to do that. Here is another way to reduce […]

]]> By: jcl https://financial-hacker.com/why-90-of-backtests-fail/#comment-108532 Tue, 15 Oct 2024 08:40:16 +0000 https://financial-hacker.com/?p=4373#comment-108532 No, the WFO period is the same for all assets.

]]> By: Tom https://financial-hacker.com/why-90-of-backtests-fail/#comment-108522 Tue, 15 Oct 2024 07:08:56 +0000 https://financial-hacker.com/?p=4373#comment-108522 In a porfolio system, would it be possible to have multiple and different number of WFO cycles (NumWFOCycles) for different symbols? f.i., symbol A has NumWFOCycles = 5, while symbol B has NumWFOCycles=10? If so, how would you do it? Thank you.

]]> By: Tom https://financial-hacker.com/why-90-of-backtests-fail/#comment-108516 Tue, 15 Oct 2024 06:43:09 +0000 https://financial-hacker.com/?p=4373#comment-108516 What I tried to do is this:
1. in the MRC file, set the CYCLES parameter to 1, i.e.,
#define CYCLES 1
2. clicked on [Train]. Training is done with original data
3. set the CYCLES parameter back to 1000, i.e.,
#define CYCLES 1000
4. clicked on [Test]

The RANDOMIZE parameter is set to SHUFFLE. Are the above steps valid for performing a MRC test? Thanks.

]]> By: jcl https://financial-hacker.com/why-90-of-backtests-fail/#comment-108209 Thu, 10 Oct 2024 15:17:57 +0000 https://financial-hacker.com/?p=4373#comment-108209 Only for this system. With other systems, it can very well matter.

The system is based on the assumption that 3-candle patterns in the training data also occur in the test data. Since shuffling replaces these pattern with random patterns, it destroys any correlation between test and training patterns. It does not matter if one of them remains original, or is also shuffled.

]]> By: Tom https://financial-hacker.com/why-90-of-backtests-fail/#comment-108150 Wed, 09 Oct 2024 17:57:38 +0000 https://financial-hacker.com/?p=4373#comment-108150 On the 1st paragraph of page 138 in the Black Book, you mentioned that during MRC run with the use of Detrend = SHUFFLE, TRAIN mode is not needed to repeat thousands of times when using WFO. The mentioned reason for this is because it does not matter whether training was done with shuffled data or with original data. Why does it not matter? Thanks.

]]> By: jcl https://financial-hacker.com/why-90-of-backtests-fail/#comment-107225 Mon, 23 Sep 2024 09:20:06 +0000 https://financial-hacker.com/?p=4373#comment-107225 Sure, a MRC with WFO training can use multiple cores and run accordingly faster.

]]> By: Tom https://financial-hacker.com/why-90-of-backtests-fail/#comment-107200 Mon, 23 Sep 2024 00:25:22 +0000 https://financial-hacker.com/?p=4373#comment-107200 Can the Montecarlo Reality Check (MRC.c) simulation time be speeded up? for example, with the use of multi-cores (NumCores = -1)? Thank you.

]]>