مقاله انگلیسی رایگان در مورد داده پایدار با ارزیابی پیش بینی مونت کارلو

Forecast evaluation methods and statistics allow for the ranking and comparison of models, but inference in time series contexts is related strongly to the degree of persistence. Local-to-unity and unit roots models are persistent processes that are considered commonly as models of interest; see Alquist and Kilian (2010), Baumeister et al. (Forthcoming), and Bernard et al. (2012) for some applications to commodity prices and macroeconomic data.

The forecast errors from persistent processes are known to follow non-standard distributions, see Kemp (1999) andPhillips (1998). Diebold and Kilian (2000) suggest using a unit root pre-test to choose a linear or first-difference forecasting model design. Their method provides some improvement over an arbitrary selection of the model structure, but the improvements rely on low-power tests for unit roots. The forecast evaluation tests for cointegrated and unit root models that were examined by Berkowitz and Giorgianni (2001) and Corradi et al. (2001) rely on nonstandard critical values for inference.

Rossi (2005) employs a Bonferroni method, based on the work of Cavanagh et al. (1995) and Stock and Watson (1988), to account for the non-standard distribution of forecast evaluation statistics. Rossi’s method focuses on a local-to-unity definition of the autoregressive parameters underlying the predictive model, which approaches the random walk forecast near the boundary even when the predictive covariates are not irrelevant. More broadly,Bonferroni bounds are known to suffer from low power, due to their conservative nature.