search

Topaceminem / DCC-GARCH

DCC GARCH modeling in Python
Apache License 2.0
84 stars 27 forks source link

About the demo #2

Open lydemo opened 4 years ago

lydemo commented 4 years ago

Hi, I run the demo on Jupyter and when it comes to the sentence: sp_model.fit(sp_return) and the sentence: jpm_model.fit(jpm_return) there exists the error: UnboundLocalError: local variable 'iprint' referenced before assignment I have no idea about such error, could you help me to solve the problem? Since I can't get the same result as yours.

Yorshh commented 3 years ago

Hi! I'm having a similar error when i try to replicate your results. In the Part of sp_model.fit(sp_return) the function is not updating the parameters of theta and just gave me back the initial values (0.005, 0.1, 0.1, 0.85). Curiosly, it seems to depend on the amount of data of each asset, because when i try with 1 year data, the values of the parameters seems to change. The same thing happend to me when i was trying to model the dcc between Bitcoin and the Dow Jones index. When i use 1 year data of Bitcoin, theta change, but when i use more than 2 years data, thetha has the initial values. Is this for overfitting? or something is wrong in the code? I would be very grateful if you can help me whit this issue :)!

Cuky88 commented 2 years ago

Hi, any updates on this? I have the same problem as @Yorshh , theta won't be updated.

YANAo121 commented 1 year ago

Hi, any updates?

Yorshh commented 1 year ago

Hi, any updates?

Hi!, at the time of my comment I didn't't find a good way to implement DCC-GARCH in Python. My solution was doing so in R where this is very well documented, you can use rugarch package. Anyways, Dynamic conditional correlation should be taken just as a reference since it is a criticized model, for example, the model doesn't take in count that two time series can be correlated but with lag. If you don't want to complicate too much you can just use rolling window correlation, or you can use another approach like using dynamic time warping to see how similar are two time series instead of using correlations, hope it helps!