Any plans to add 1)Regression modeling (Simple (SLR) and multiple linear regression (MLR) ), 2) step-wise regression and 3) Collinearity examination or test among explanatory variables?
Excel has support for simple and multiple linear regression in Analysis pack add-in. Unfortunately, the output is a bit rigid to a predefined table output in your worksheet (plain numbers, no formulas or link back to source data).
Performing manually a series of regression models using the analysis pack add-in on the same sheet (e.g. back-testing, step-wise regression, etc.) is very difficult, confusing to impossible. Furthermore, there is no support for col linearity test or structure break (aka homogeneity test).
Furthermore, updating the models as new data becomes available requires re-doing the whole analysis (i..e new output tables, diagnosis, forecast, etc.).
In practice, an analyst starts with a dependent variable (i.e. Y) and a set of potential explanatory variables sets. The end result is a regression model that best explains the variation in Y by exploiting information in X.
The Analyst.modeler wishes to verify the stability of the model over their sample to ensure reliable forecast. Furthermore, occasionally, analyst wishes to update their model and forecast as new data becomes available.
The regression model is the stable of all models in practice, and we need to make the process very simple, yet thorough to diagnose data or model's nuisances as much as possible.
Would it possible to include the lagged version of the input variables (for time series)? This will be great, especially with step-wise regression to narrow the optimal set of input variables that explain the variation in the dependent time series variable.
How does this relates to vector autoregressive (VAR) or vector error correction model (VECM)?
Comments
Excel has support for simple and multiple linear regression in Analysis pack add-in. Unfortunately, the output is a bit rigid to a predefined table output in your worksheet (plain numbers, no formulas or link back to source data).
Performing manually a series of regression models using the analysis pack add-in on the same sheet (e.g. back-testing, step-wise regression, etc.) is very difficult, confusing to impossible. Furthermore, there is no support for col linearity test or structure break (aka homogeneity test).
Furthermore, updating the models as new data becomes available requires re-doing the whole analysis (i..e new output tables, diagnosis, forecast, etc.).
In practice, an analyst starts with a dependent variable (i.e. Y) and a set of potential explanatory variables sets. The end result is a regression model that best explains the variation in Y by exploiting information in X.
The Analyst.modeler wishes to verify the stability of the model over their sample to ensure reliable forecast. Furthermore, occasionally, analyst wishes to update their model and forecast as new data becomes available.
The regression model is the stable of all models in practice, and we need to make the process very simple, yet thorough to diagnose data or model's nuisances as much as possible.
This is done, and is available in the official release of NumXL 1.60 (APACHE)
Would it possible to include the lagged version of the input variables (for time series)? This will be great, especially with step-wise regression to narrow the optimal set of input variables that explain the variation in the dependent time series variable.
How does this relates to vector autoregressive (VAR) or vector error correction model (VECM)?
Please sign in to leave a comment.