Bayesian model - how to emphasize later observations? I have a very general question, any links to relevant papers or which book I should consult should suffice.
So, let's say I've got a Bayesian model (for example) to predict the outcome of a soccer game in the current season. I have some data for the current season (since not all games have been played), and then I have the data from the previous couple of seasons. 
Is there a method in Bayesian statistics whereby you would develop a model that provides more weight for the current season results compared to the results from the previous seasons? 
Appreciate the help!
 A: This sounds like a Hidden Markov Model. You have some data about the state of your system in the past (data = soccer game results, state = how likely teams are to win (e.g.)). Then you have some data about the state of the system in the present. You assume that the system's current state is somewhat similar to its state in the recent past. If you could somehow formalize this temporal relationship ($p(x_t|x_{t-1})$, where $x_t$ is the state of the system at time $t$), you could then use Recursive Bayesian estimation to infer the current state of the system from both present and previous data. This gives you a better estimate of the current state (i.e. how good the different soccer teams currently are), and that then allows you to make better predictions. 
This automatically puts more weight on the more recent data, as long as your state transition model allows for some stochasticity. That is, as long as you assume that previous states are not perfectly predictive of current states, then (all else being equal, e.g. equal noise) the more recent data is always a statistically better source of information, which will therefore be weighted more heavily.
