Aggregating strategies for longterm forecasting
[edit]
Proceedings of the Seventh Workshop on Conformal and Probabilistic Prediction and Applications, PMLR 91:6382, 2018.
Abstract
The article is devoted to investigating an application of aggregating algorithms to the problem of the longterm forecasting. We examine the classic aggregating algorithms based on the exponential reweighing. For the general Vovk’s aggregating algorithm we provide its probabilistic interpretation and its generalization for the longterm forecasting. For the special basic case of Vovk’s algorithm we provide two its modifications for the longterm forecasting. The first one is theoretically close to an optimal algorithm and is based on replication of independent copies. It provides the timeindependent regret bound with respect to the best expert in the pool. The second one is not optimal but is more practical (explicitly models dependencies in observations) and has $O(\sqrtT)$ regret bound, where $T$ is the length of the game.
Related Material


