Non stationary data - Good model parameters

[andrewcztrack]

Hi @ourownstory !!!

I hope your well.

I am trying to use the model on non stationary stock data. I understand this is a hard problem but i wondering with the right understanding, can good estimates be calculated. Below is my attempt. Does the pricing data need to be transformed first in order to be trained correctly. Should i employ a box cox transformation before training on the model? Is the parameters that I am using to fit the model good estimates?

Kind regards and thanks,
Andrew

import pandas as pd 
from neuralprophet import NeuralProphet
import numpy as np
import matplotlib.pyplot as plt
from scipy import stats 


%matplotlib inline
import quantstats as qs

# extend pandas functionality with metrics, etc.

import yfinance as yf
df = yf.download("VT", start="2012-01-01", end="2021-04-30")['Adj Close']
#fitted_data, fitted_lambda = stats.boxcox(df) 
#df2=pd.concat([new, df], axis=1)


df=df.reset_index()

df=df.rename(columns={'Date':'ds',  'Adj Close' :'y'})


# model = NeuralProphet() if you're using default variables below.
model = NeuralProphet(
    growth="linear",  # Determine trend types: 'linear', 'discontinuous', 'off'
    changepoints=None, # list of dates that may include change points (None -> automatic )
    n_changepoints=5,
    changepoints_range=0.8,
    trend_reg=0,
    trend_reg_threshold=False,
    yearly_seasonality="auto",
    weekly_seasonality="auto",
    daily_seasonality="auto",
    seasonality_mode="additive",
    seasonality_reg=0,
    n_forecasts=1,
    n_lags=0,
    num_hidden_layers=0,
    d_hidden=None,     # Dimension of hidden layers of AR-Net
    ar_sparsity=None,  # Sparcity in the AR coefficients
    learning_rate=None,
    epochs=250,
    loss_func="Huber",
    normalize="auto",  # Type of normalization ('minmax', 'standardize', 'soft', 'off')
    impute_missing=True
)

metrics = model.fit(df, validate_each_epoch=True, freq="D")
future = model.make_future_dataframe(df, periods=80, n_historic_predictions=len(df)) 
forecast = model.predict(future)

[ourownstory]

Hi @andrewcztrack

Did you get any pleasing results?
A boxcox transform can help.
Y could also add lagged regressors.

I would recommend leaving unused options blank (args you are not manually setting)