Has anyone actually built a robust sub-1h trading model?
Has anyone actually built a robust sub-1h trading model?
Loading saved threads...
Adrian Velai · External communityPost link
External question — Quantitative Finance Stack Exchange
Author: Adrian Velai
Original post: https://quant.stackexchange.com/questions/85868
License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0/
Adaptation: HTML converted to plain text; contact email addresses removed.
I keep finding the same breakpoint in systematic trading research:
Below 1h, apparent edge tends to disappear once you include realistic costs, slippage, walk-forward testing, parameter perturbation and different training cutoffs.
From 1h upwards, I have been able to obtain models that remain materially more stable under those same tests.
So I am curious about the opposite case:
Has anyone here built a 5m–30m model that remained genuinely robust out of sample after realistic execution costs?
If yes, what was the key difference?
Was it mainly:
richer data than OHLCV,
order-flow/microstructure features,
much lower trade frequency,
execution modeling,
or a fundamentally different validation approach?
I am not interested in impressive in-sample Sharpe ratios. I mean models that still look good after serious robustness testing.
Quote
Report
Post Reply
Quoted from Forex.com.bd-Editorial External question — Quantitative Finance Stack Exchange Author: Adrian Velai Source score (net votes, not local likes): 0 Original post: https://quant.stackexchange.com/questions/85868 License: CC BY-SA 4.0 — https://creativecommons.org/licenses/by-sa/4.0/ Adaptation: HTML converted to plain text; contact email addresses removed. I keep finding the same breakpoint in systematic trading research: Below 1h, apparent edge tends to disappear once you include realistic costs, slippage, walk-forward testing, parameter perturbation and different training cutoffs. From 1h upwards, I have been able to obtain models that remain materially more stable under those same tests. So I am curious about the opposite case: Has anyone here built a 5m–30m model that remained genuinely robust out of sample after realistic execution costs? If yes, what was the key difference? Was it mainly: richer data than OHLCV, order-flow/microstructure features, much lower trade frequency, execution modeling, or a fundamentally different validation approach? I am not interested in impressive in-sample Sharpe ratios. I mean models that still look good after serious robustness testing.
Checking account access…