In your experience, how do you know or what do you need to check in order to declare econometric data as bad (univariate & multivariate cases).
No honestly obtained data is bad. My ideas are expressed in the attached. D, Booth
No honestly obtained data is bad. My ideas are expressed in the attached. D, Booth
To the excellent answer by David Eugene Booth I would like to add that models (univariate or multivariate) can be "bad" if they are fitted to the data by disregarding their underlying assumptions. Another criterion to distinguish "bad" from "good" model is to ask if the model is predicting anything definite, accurate, consistent and/or useful?
DATA cannot be classified as "good" or "bad." Data is "data". How you use data and for what purpose do you use the data may give a specific result. The judgment should be pass on to evaluate the result, not the data.
OUTLIER DETECTION may be detected through extreme value theory (EVT). If the data set contains significant amount of extreme values (outliers), it might take a completely unique type of distribution of its own, i.e.rechet, Weibull, or Gumber, see below for how to calculate tail index to classify the type of distribution.
http://www.unige.ch/ses/dsec/static/gilli/evtrm/GilliKelleziCE.pdf
UNIVARIATE & MULTIVAIRATE has nothing to do with outliers. How many variates you proposed for a model does not affect the nature of the data. Outliers is an intrinsic characteristic of the data set. Variates deals with how many variables do you use for your modeling, i.e. x1, x2, etc.
Some useful links:
https://archive.siam.org/meetings/sdm10/tutorial3.pdf
https://en.wikipedia.org/wiki/Anomaly_detection
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics