Hello, the oscillator is as follows.
here, s = diff.std(), this std uses full data series, isn't it a leakage of future data in early days?

compute the oscillator

def osc(prices, fast=32, slow=96, scaling=True):
"""
oscillator

Args:
    prices: a dataframe of prices
    fast: fast moving average factor, e.g. 32
    slow: slow moving average factor, e.g. 96
    scaling: true/false. If true scales with the standard deviation of the signal
    Strictly speacking this step is forward looking.

Returns:
    oscillator
"""
diff = prices.ewm(com=fast - 1).mean() - prices.ewm(com=slow - 1).mean()
if scaling:
    s = diff.std()
else:
    s = 1

return diff / s

see
https://stackoverflow.com/questions/76158826/show-statistics-average-stddev-min-max-for-visible-data-with-plotly

Thank you for installing the Korbit AI Mentor!

To get started, please register an account on our website by authenticating with GitHub authentication. If you already have an account with us, please make sure it is linked to your GitHub account.

Once registered, we'll start reviewing your pull requests and you can kick of a scan manually from our website.

Hello, I ran experitment5 against p_hashed.csv and p_ob.csv and get very different results. the result of p_hashed is very bad and the result of p_ob is very good. Just wonder what the difference between p_hashed.csv and p_ob.csv is?

Hi Dr. Schmelzer,
First of all thanks for sharing this work, and I am following the code line by line where I found a potential issue(which might be my lack of understanding),
so we have this right now:

portfolio = Portfolio(prices=prices, position=prices.apply(f, slow=slow.value, fast=fast.value, vola=vola.value, clip=winsor.value))

where the position passed to Portfolio Class varies quite violently, I'd rather have it like this:

position=prices.apply(f, slow=slow.value, fast=fast.value, vola=vola.value, clip=winsor.value)

My Suggestion Rescale the position as:

plus if would want to avoid leverage

position = position.apply(lambda x: x/x.abs().sum(), axis=1)
and then pass this to
portfolio = Portfolio(prices=prices, position=position)

Basically not just experiment 3 but in all the schemes. Do let me know, I might be terribly wrong.

Hi Dr. Schmelzer,
I have learned so much from your work, thank you very much.
I have a question in Experiment5, you say in conclusions
"Possible to reflect trading costs in objective with regularization terms (Ridge, Lars, Elastic Nets, ...).

Could you give a small example of this regularization term?