Improving Market Regime Detection: A New Approach Combining Markov-Switching Models and Adaptive Moving Averages
deepseek link not working for me interested to read more about the technical details
Me too
Is there ipynb file for this?
It’s 🔥
Would be great to the full ipynb
first of all, thank you for sharing your code and your overall efforts. I read quickly, so apologies if i missed something. Some remarks:
- it would be best to look for regimes in a walk forward manner
- in-sample backtest plus optimised KAMA say little about ability to generalise to future states
- you imply it in your text: we need features to avoid jumping from one state to another, and KAMA is just one price feature
- in summary: walk forward, some more relevant features (prices, but of competing asset classes or of even better of risk premia)
I know, you won’t make all of that public, but still worth mentioning
Doesn't
`filtr = gamma * np.std(kama_diff)`
in `get_kama` suffer from look-ahead bias?
I think that
`kama_diff.rolling(window=n_window, min_periods=1).std()`
would be a suitable fix.
What do you think?
Nice write up
Cool stuff would be useful for filters on trading systems.
Great piece
deepseek link not working for me interested to read more about the technical details
Me too
Is there ipynb file for this?
It’s 🔥
Would be great to the full ipynb
first of all, thank you for sharing your code and your overall efforts. I read quickly, so apologies if i missed something. Some remarks:
- it would be best to look for regimes in a walk forward manner
- in-sample backtest plus optimised KAMA say little about ability to generalise to future states
- you imply it in your text: we need features to avoid jumping from one state to another, and KAMA is just one price feature
- in summary: walk forward, some more relevant features (prices, but of competing asset classes or of even better of risk premia)
I know, you won’t make all of that public, but still worth mentioning
Doesn't
`filtr = gamma * np.std(kama_diff)`
in `get_kama` suffer from look-ahead bias?
I think that
`kama_diff.rolling(window=n_window, min_periods=1).std()`
would be a suitable fix.
What do you think?
Nice write up
Cool stuff would be useful for filters on trading systems.
Great piece