Skip to main content

New answers tagged

0 votes

Strange increasing in $R^2$ when MAE and RMSE worsened for OLS

I think I see what happened. You regress the true values on the predictions, meaning that your linear regression $R^2$ value is equal to the squared Pearson correlation between the true and predicted ...
Dave's user avatar
  • 62.6k
0 votes

ARIMAX-GARCH flattens when using daily return, but not level

Your eyes may be tricking you. What matters is the vertical distance between the forecast and the corresponding realization. In the first graph, the vertical distance is quite hard to judge. Also, the ...
Richard Hardy's user avatar
0 votes
Accepted

Approach to Handling Stationarity in Multi-dimensional Time Series Forecasting with AutoARIMA

Generally speaking, you don't need to make your data stationary when using ARIMA. That's the idea of the "I" (integrated) which will differentiate your series before training. You would need ...
Philippe Ostiguy's user avatar
1 vote

Paradox of Brier skill score of perfectly calibrated output?

The Brier skill score is designed to assess the forecast skill of all possible forecasts and data generating processes. As Dave writes, it equals 1 if, and only if the prediction is perfect, i.e. ...
picky_porpoise's user avatar
5 votes

Paradox of Brier skill score of perfectly calibrated output?

Brier skill score is equal to $1-\dfrac{BS}{BS_{ref}}$, where the numerator is the model Brier score and the denominator is the Brier score of a reference model (a baseline model or some kind of ...
Dave's user avatar
  • 62.6k
0 votes

Why the differenced at lag 12 time series of a SARIMA(0,0,0)(0,1,1)_12 model follow the MA(1) pattern with step 12?

SARIMA(0,0,0)(0,1,1)12 is obtained by cumulatively summing SARIMA(0,0,0)(0,0,1)12 for every of the 12 seasonal periods. Differencing undoes the cumulative summation. I hope it is clear why the latter ...
Richard Hardy's user avatar
0 votes

Rolling forecasts where horizon is larger than step-size

You should produce your forecasts for the horizon for which they are necessary. If you need forecasts for one week ahead, then that's fine. How often you produce such forecasts is completely ...
picky_porpoise's user avatar
1 vote

Contradictory Sources on Seasonality being a nonstationarity

This may be neither rigorous enough nor detailed enough to be an answer you are looking for, but let me offer a brief take on the matter. Seasonal time series can be either stationary or nonstationary....
Richard Hardy's user avatar
2 votes

Forecasting a series that comes with uncertainty

The answer depends much on what you want to forecast and also on the plausible models generating the data. For the first aspect you may want to forecast the spatial average at a future point in time ...
Yves's user avatar
  • 5,398
0 votes

Multivariate time series with additional features

The {mvgam} R package was designed specifically for these kinds of problems. It handles both univariate and multivariate time series, and is set up to allow for ...
Nicholas Clark's user avatar
2 votes

How to fit a GAM with double seasonality to a daily time series? (mgcv package)

As the comment from @user11852 suggests, you can incorporate multiple forms of seasonality (and how those seasonalities may change over time) using the cyclic cubic regression basis in the {...
Nicholas Clark's user avatar
2 votes

Forecasting a series that comes with uncertainty

It sounds like you may have entire probability distributions you would like to forecast. This is a case of functional-data-analysis. We don't have a lot on functional data forecasting; this thread may ...
Stephan Kolassa's user avatar
6 votes

Re-selecting best sales forecasting model each month. Is this overfitting?

The obvious way to figure out whether this works (as compared to other strategies) is to compare the forecasts vs. the alternative ideas afterwards (or initially just looking backwards). A few things ...
Björn's user avatar
  • 32.3k
6 votes

Re-selecting best sales forecasting model each month. Is this overfitting?

Forecasting is a mixture of science, engineering and art. There are definitely ways that this approach can go wrong... but equally, there are ways in which many other "reasonable" approaches ...
Stephan Kolassa's user avatar
1 vote

How to deal with zeros when using NBEATS to forcast demand?

I've come across the Croston model, but I'm not looking for an average; rather, I want to predict the number of people who will want a particular room on a given day. At very granular levels (here: ...
Stephan Kolassa's user avatar
0 votes

Non-Continuous Time Series Forecasting

If you're working with a statistical package for time series forecasting that uses internally state-space models representation, then it'll be automatically handled by the software. For instance, ...
Aksakal's user avatar
  • 61.4k
0 votes

Is a larger beta weight a better predictor than a high t-statistic?

With respect to path models. The standard answer here, use the standardized beta is a rote answer which in practice can be far less optimal. Related to this and pertinent to this discussion is that, ...
Scott U's user avatar
0 votes

auto_arima straight line prediction python

Maybe you need a slight change in your code as follows: ...
Reza Nadimi's user avatar

Top 50 recent answers are included