Generalization Error Bounds for Time Series

摘要: In this thesis, I derive generalization error bounds — on the expected inaccuracy of predictions for time series forecasting models. These allow forecasters to select among competing models, and declare that, with high probability, their chosen model will perform well without making strong assumptions about data generating process or appealing asymptotic theory. Expanding upon results from statistical learning theory, demonstrate how these techniques can help choose models which behave under uncertainty. also show estimate β-mixing coefficients dependent so that my be used empirically. use bound explicitly evaluate different predictive volatility IBM stock a standard set macroeconomic variables. Taken together control fixed growing memory.