使用 ETS 进行预测会产生意想不到的常数值

Question

我想根据一系列 15 个数据点进行预测（我知道样本很小）

数据列表：

acper <- c(0.17437418,0.157034518,0.156229944,0.165366982,0.17212133,0.205634804,0.177709816,0.162522795,0.165439735,0.147619365,0.144451531,0.151751211,0.129296274,0.129692449,0.118168555)

现在我想用 ets 做预测

ets.model <- ets(acper)
forecast(ets.model, h=3)
plot(forecast(ets.model, h=3))

我对接下来的所有年份都有相同的预测

Point Forecast      Lo 80     Hi 80      Lo 95     Hi 95
16      0.1181686 0.09764478 0.1386923 0.08678015 0.1495570
17      0.1181686 0.08914355 0.1471936 0.07377865 0.1625585
18      0.1181686 0.08262033 0.1537168 0.06380224 0.1725349
19      0.1181686 0.07712100 0.1592161 0.05539174 0.1809454
20      0.1181686 0.07227600 0.1640611 0.04798195 0.1883552
21      0.1181686 0.06789578 0.1684413 0.04128298 0.1950541
22      0.1181686 0.06386775 0.1724694 0.03512264 0.2012145
23      0.1181686 0.06011855 0.1762186 0.02938874 0.2069484
24      0.1181686 0.05659723 0.1797399 0.02400334 0.2123338
25      0.1181686 0.05326668 0.1830704 0.01890970 0.2174274

为什么我总是得到0.1181686？这对我来说似乎无效。

Answer 1

指数平滑为较新的样本赋予更大的权重，为较旧的样本赋予较小的权重，而移动平均值（R 中的 ARIMA 模型）将为所有观测值赋予相同的权重（这可能更适合您的样本）。由于你的样本太小了，好像只是取最后一个值向上取整。

Answer 2

首先转换为ts：

library(forecast)

acper <- c (0.17437418,0.157034518,0.156229944,0.165366982,0.17212133,0.205634804,0.177709816,0.162522795,0.165439735,0.147619365,0.144451531,0.151751211,0.129296274,0.129692449,0.118168555)

ets.model <- ets(ts(acper, start = c(2000, 1), freq = 12), model = "ZAN")
forecast(ets.model, h=3)
#Point Forecast      Lo 80     Hi 80      Lo 95     Hi 95
#Apr 2001      0.1261377 0.10238661 0.1498888 0.08981353 0.1624619
#May 2001      0.1227179 0.09840193 0.1470338 0.08552986 0.1599059
#Jun 2001      0.1192980 0.09442959 0.1441664 0.08126504 0.1573310
plot(forecast(ets.model, h=3))

使用 ETS 进行预测会产生意想不到的常数值

Forecasting with ETS yields unexpected constant value

r

forecasting