forecasters seem to have (at least) three separate mind sets. Are there more? Is one “best”?
a) Simulation modellers: operational weather forecasters, those that send space probes to other planets... These guys tend to believe there really are mathematical laws of physics that describe (some would say “govern”) reality and that their models tend to capture those laws if imperfectly. These “realistic modellers” aim to be more interested in reality than in the maths of their models. Their models are often deterministic, but they admit “uncertainty” (imprecision) in initial conditions and parameter values.
b) Statistical modellers: traditional “time series analysis” statisticians, “professional business forecasters”... These guys are more likely to fit a stochastic time series model, and may be more interested in the properties of the model class from which the model is selected than the particular phenomena being forecast. They consider stochastic models almost exclusively, and deprecate the use of models with many parameters. Those that actually forecast often forecast many different systems with the same model (class).
c) Empirical modellers: seasonal forecasters, forecasts of small systems... These guys may think like the physicists, but accept that (today’s) simulation models do not have the fidelity to provide useful forecasts for the phenomena of interested. They aim to avoid models with (too) many parameters. They may include random terms, but often with a very different motivation from either of the other groups.
Is there a better way to describe or group model based forecasters? Are there more than three? Do they really believe different things? Is one of them better?
Is there at least one person on jref that holds each view?
a) Simulation modellers: operational weather forecasters, those that send space probes to other planets... These guys tend to believe there really are mathematical laws of physics that describe (some would say “govern”) reality and that their models tend to capture those laws if imperfectly. These “realistic modellers” aim to be more interested in reality than in the maths of their models. Their models are often deterministic, but they admit “uncertainty” (imprecision) in initial conditions and parameter values.
b) Statistical modellers: traditional “time series analysis” statisticians, “professional business forecasters”... These guys are more likely to fit a stochastic time series model, and may be more interested in the properties of the model class from which the model is selected than the particular phenomena being forecast. They consider stochastic models almost exclusively, and deprecate the use of models with many parameters. Those that actually forecast often forecast many different systems with the same model (class).
c) Empirical modellers: seasonal forecasters, forecasts of small systems... These guys may think like the physicists, but accept that (today’s) simulation models do not have the fidelity to provide useful forecasts for the phenomena of interested. They aim to avoid models with (too) many parameters. They may include random terms, but often with a very different motivation from either of the other groups.
Is there a better way to describe or group model based forecasters? Are there more than three? Do they really believe different things? Is one of them better?
Is there at least one person on jref that holds each view?
via JREF Forum http://forums.randi.org/showthread.php?t=263952&goto=newpost
Aucun commentaire:
Enregistrer un commentaire