Time series generation - returns or prices.

[andrewcz]

Hi @RudyMorel I hope you're well. A couple of brief questions.

• Is this designed for returns or prices (or either), as returns are mostly stationary.
• how would you generate many time series to understand the distribution of the process. Cheers, Andrew

[RudyMorel]

Hi @andrewcz,

• The model implemented in this repository is stationary. Thus, the log-returns (which can be assumed stationary at certain granularity) should be considered.
• to generate multiple time-series you can use the "S" argument of the function "generate" from frontend.py.

I plan to release a tutorial.ipynb to demonstrate the usage of these functions. Currently, code examples can be found in testing.ipynb

Thank you fo your questions. Rudy

[andrewcz]

Thank you @RudyMorel! Will give it a try.

On Wed, 7 Feb 2024 at 8:17 am, Rudy Morel @.***> wrote:

Hi @andrewcz https://github.com/andrewcz,

• The model implemented in this repository is stationary. Thus, the log-returns (which can be assumed stationary at certain granularity) should be considered.
• to generate multiple time-series you can use the "S" argument of the function "generate" from frontend.py.

I plan to release a tutorial.ipynb to demonstrate the usage of these functions. Currently, code examples can be found in testing.ipynb

Thank you fo your questions. Rudy

— Reply to this email directly, view it on GitHub https://github.com/RudyMorel/scattering_spectra/issues/8#issuecomment-1930765691, or unsubscribe https://github.com/notifications/unsubscribe-auth/AARQUD3ATW3IBJHAEGGSSHTYSKMXJAVCNFSM6AAAAABCXJUDACVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMZQG43DKNRZGE . You are receiving this because you were mentioned.Message ID: @.***>

[RudyMorel]

Hi @andrewcz,

In the last commit I added a tutorial.ipynb file.

Also, coming back to your question, I realize we may have different definitions of the word "returns". There are four quantities:

• the prices
• the returns (i.e. price-increments)
• the log-prices
• the log-returns (i.e. log-price increments)

A Scattering Spectra model accepts both log-prices or log-returns. More generally, it accepts stationary time-series (e.g. log-returns) or time-series which are not necessarily stationary but whose increments are (e.g. log-prices).

Rudy