La base de datos para el siguiente análsisi es extraÃdo de Yahoo Finance (a través de la técnica scraping) y trata sobre el precio de las acciones de McDonalds.
#install.packages('quantmod') # librerÃa para hacer scraping
#install.packages('tseries')
#install.packages('timeSeries')
#install.packages('forecast')
#install.packages('xts')
#install.packages('ggplot2')
library(quantmod)
## Loading required package: xts
## Loading required package: zoo
##
## Attaching package: 'zoo'
## The following objects are masked from 'package:base':
##
## as.Date, as.Date.numeric
## Loading required package: TTR
## Registered S3 method overwritten by 'quantmod':
## method from
## as.zoo.data.frame zoo
library(tseries)
library(timeSeries)
## Loading required package: timeDate
##
## Attaching package: 'timeSeries'
## The following object is masked from 'package:zoo':
##
## time<-
library(forecast)
library(xts)
library(ggplot2)
MCD <- getSymbols('MCD', src='yahoo', from = as.Date("2021-01-01"),to=as.Date("2023-07-27"), auto.assign = FALSE)
Graficando la serie
chartSeries(MCD, name="MCD", subset="last 6 months", theme=chartTheme("white"))
Datos de yahoo finance:
data_MCD <- data.frame(MCD, tiempo = as.Date(rownames(data.frame(MCD))))
head(data_MCD)
## MCD.Open MCD.High MCD.Low MCD.Close MCD.Volume MCD.Adjusted
## 2021-01-04 214.49 214.72 208.22 210.22 4055400 198.8033
## 2021-01-05 210.18 211.95 209.62 211.48 2576100 199.9949
## 2021-01-06 211.30 211.71 209.03 211.00 3083400 199.5410
## 2021-01-07 213.22 213.22 210.56 211.98 3142000 200.4678
## 2021-01-08 212.90 216.12 212.23 215.87 2639100 204.1465
## 2021-01-11 215.09 216.12 213.12 214.23 2545400 202.5956
## tiempo
## 2021-01-04 2021-01-04
## 2021-01-05 2021-01-05
## 2021-01-06 2021-01-06
## 2021-01-07 2021-01-07
## 2021-01-08 2021-01-08
## 2021-01-11 2021-01-11
attach(data_MCD)
Separando la serie close:
base1 = data.frame(tiempo, MCD.Close)
names (base1) = c("tiempo","MCD")
base1 <- na.omit(base1) #eliminando datos ominitidos "NA"
#base1
head(base1, n = 10)
## tiempo MCD
## 1 2021-01-04 210.22
## 2 2021-01-05 211.48
## 3 2021-01-06 211.00
## 4 2021-01-07 211.98
## 5 2021-01-08 215.87
## 6 2021-01-11 214.23
## 7 2021-01-12 211.60
## 8 2021-01-13 212.09
## 9 2021-01-14 208.50
## 10 2021-01-15 209.91
Graficando la serie
ggplot(base1, aes(x = tiempo, y = MCD)) + geom_line() + geom_smooth(se = FALSE)+ labs(title = "Precio de las acciones de McDonalds", x = "Fecha", y = "Precio / Acción")
## `geom_smooth()` using method = 'loess' and formula = 'y ~ x'
MCD_ma = ts(na.omit(base1$MCD), frequency=30)
decomp = stl(MCD_ma, s.window="periodic")
deseasonal_base1 <- seasadj(decomp)
plot(decomp)
adf.test(MCD_ma, alternative = "stationary")
##
## Augmented Dickey-Fuller Test
##
## data: MCD_ma
## Dickey-Fuller = -3.4784, Lag order = 8, p-value = 0.04438
## alternative hypothesis: stationary
Contraste de hipótesis:
pp.test(MCD_ma, alternative = "stationary")
##
## Phillips-Perron Unit Root Test
##
## data: MCD_ma
## Dickey-Fuller Z(alpha) = -22.836, Truncation lag parameter = 6, p-value
## = 0.03988
## alternative hypothesis: stationary
Contraste de hipótesis:
Las ACF proporcionan información sobre cómo una observación influye en las siguientes.
Acf(MCD_ma, main='')
Pacf(MCD_ma, main='')
Para realizar un modelo ARIMA, la serie temporal debe ser estacionaria. Para conseguir esta estacionariedad, la diferenciaremos.
MCD_d1 = diff(deseasonal_base1, differences = 1)
plot(MCD_d1)
Para comprobar que la serie es, efectivamente, estacionaria, hacemos de nuevo el test aumentado de Dickey-Fuller.
adf.test(MCD_d1, alternative = "stationary")
## Warning in adf.test(MCD_d1, alternative = "stationary"): p-value smaller than
## printed p-value
##
## Augmented Dickey-Fuller Test
##
## data: MCD_d1
## Dickey-Fuller = -8.3573, Lag order = 8, p-value = 0.01
## alternative hypothesis: stationary
modeloarima<-auto.arima(MCD_ma, seasonal=FALSE)
modeloarima
## Series: MCD_ma
## ARIMA(0,1,0)
##
## sigma^2 = 6.935: log likelihood = -1535
## AIC=3071.99 AICc=3072 BIC=3076.46
tsdisplay(residuals(modeloarima), lag.max=10, main='(0,1,0) Model Residuals')
El modelo ARIMA es correcto, pues tiene una tendencia creciente y su modelo es 0,1,0. Donde en parte nos indica que no hay errores o innovaciones (0)
prediccion <- forecast(modeloarima, h=30)
plot(prediccion)
tail(prediccion$mean,30)
## Time Series:
## Start = c(22, 15)
## End = c(23, 14)
## Frequency = 30
## [1] 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75
## [11] 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75
## [21] 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75 291.75