Hi all, I'd like some suggestions about this. It seems to me that when you estimate the posterior probability, your priors are given equal weight (in the standard bayesian model), but in reality, priors ought to be given weights depending on how low ago they are, but how do account for this in your posterior estimates? Has anybody got any academic papers on the topic from either within sports gambling or outside sports gambling (social science surely has some answers, eg. more recent election polls are more important than less recent ones in predicting election outcomes anybody come across any papers?)
-
-
My understanding of Bayesian is lackluster at best. But from the question posed, if recent events were significantly different from past ones, enough that weighing them more would change things, then why include the past results at all? A significant change in probabilities would indicate a different situation.
An example I would be more familiar with:
Say when Oil has risen in the past, gold also had a 40% chance of rising in the same day, but for the last three years, gold has only risen 10% of the time when Oil does. Why would I include results past 3 years? The underlier is a different situation now vs. in the past. It would throw off my calculations if I included too much past data.
Sorry if I misunderstood the question, but if the underlier is changing significantly, why include data from long ago?Comment -
Actually I'm very sorry, I've now realized what I'm asking is pointless... Not sure why I made the thread actually
.
Comment