|
Time series analysis and forecasting of economic statistics plays an important role, but so far, most of the literature only for period of economic time series are discussed, for non-economic time series involves less time, however, non-period of economic time series is widespread The. For example, a monthly warehouse inventory, a company in 1997, the monthly number of employees at the end of the statistics, which are de facto economic time series; addition, the average time series and the relative number of time series, such as The average wage is the average index time series, the wage index is a relative index time series. Since the non-period of economic time series analysis, including de facto economic time series analysis, the average number of economic time series analysis, the relative number of economic time series analysis, involves a broader. This article focuses equally spaced stationary point in AR (p) economic time series, for example, to study the numerical characteristics of the model initial identification, parameter estimation and prediction. Several other economic time series analysis can be a similar discussion. For convenience, the timing of said time interval equal to the point of economic time series economic time series. As indicators of economic time series point value added of no practical significance, so the point of economic time series, its digital signature can not use the previous formula. This paper first introduces the de facto economic time series characteristics of the basic concepts and figures, and their numerical characteristics of the sample properties are discussed. There are the following results: 1) define the binary method using both sample mean, since the sample covariance function is defined as: The sample autocorrelation function and the sample partial autocorrelation function has changed accordingly. 2) the nature of the sample numerical characteristics a) x is an unbiased estimate of the mean μ and consistent estimates. b) the sample from the sample covariance function γk autocorrelation function pk pk respectively γk with the asymptotically unbiased estimator and tim \· Tim Er = to a. Team 4) c) Let kl eye from AR (n) model, {a for the independent and identically distributed white noise sequence, Ea; 二 0, Ea7 two. '<. Eat 1 (0.5) tim 9. ; = P. ; A. . = L, 2,, · ·, k, k> 1 (0.6) and when k> p, its partial correlation function is to write it in, and the random vector (grasping pair H,. Published, · ·, grab w +.,. +.) asymptotically Chuan 0 people) and several of them as a unit matrix of order m. > 1 for any given positive integer. Section II is the initial identification of the model. Ie the initial recognition model categories, according ARb) model partial autocorrelation function of the determined order Imperial censored. Section III is the parameter estimation, mainly moment estimation, least squares estimation with maximum likelihood estimation method to estimate AWP) model parameters. Mainly has the following results: M) ARp) Moment estimation of model parameters ibM = YJ'R. Where, F. Since the sample covariance function matrix, Rn = (, ..., yr) 'D b) AR ...) model parameter least squares estimation iLS two Bu ... JJRL ... which, I7 B \\ UJ 7 working ti) \LLI, - 1) \\ _I 1LL people 1-u \\ called \\\ Ding L \\ corpse one on) 7L \\ corpse L) \* = two> (a) (a) [magic. -.. j. -.. '? fly oh-\(MIM \\ l A a 17 WML \\\ Ltool \\ A on 0d2ld22 '\nn \\ \=>. Js + l-. JsOS. , JSp (0.m) s. 1 + J Z By comparison, when n is large, these three estimates less. As moments estimate of up easily, so commonly moment estimation. Section IV presents the AR * model order selection, mainly used AIC criterion for order determination. Section V AR {p) sequences forecasts. The main results are as follows: Call to set up a ... ·. x-N, N; play) is zero mean and variance of the random vector limited, into the BU) as dependent *;, n *, ...,. ; -. ·) Resulting optimal prediction, then azelaic small) = E (X soil + Jung Boc, ..., X ± a.) Bu) AR a) sequences recursive prediction formula corpse Tf (1 \long (2) = plx; (1) + pZx?
|