---
title: "time_series"
author: "Kathleen Durant"
date: "October 8, 2018"
output: html_document
---

```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
```

## R Markdown


```{r loadlib}
library(tidyverse)      # data manipulation and visualization
library(lubridate)      # easily work with dates and times
library(fpp2)           # working with time series data
library(zoo)            # working with time series data
```

Moving Average
Choose a number of nearby points and average them to estimate the trend. When calculating a simple moving average, it is beneficial to use an odd number of points so that the calculation is symmetric. 
Check out the description of the rollmean function

```{r simple}
savings <- economics %>%
  select(date, srate = psavert) %>%
  mutate(srate_ma01 = rollmean(srate, k = 13, fill = NA),
         srate_ma02 = rollmean(srate, k = 25, fill = NA),
         srate_ma03 = rollmean(srate, k = 37, fill = NA),
         srate_ma05 = rollmean(srate, k = 61, fill = NA),
         srate_ma10 = rollmean(srate, k = 121, fill = NA))
```
```{r plot_data}
savings %>%
  gather(metric, value, srate:srate_ma10) %>%
  ggplot(aes(date, value, color = metric)) +
  geom_line()
```

As the number of points used for the average increases, the curve becomes smoother and smoother. Choosing a value for k is a balance between eliminating noise while still capturing the data's true structure. 

```{r}
savings %>%
  gather(metric, value, srate_ma01:srate_ma10) %>%
  group_by(metric) %>%
  summarise(MSE = mean((srate - value)^2, na.rm = TRUE),
            MAPE = mean(abs((srate - value)/srate), na.rm = TRUE))
```

As the number of points increases, so does the error. It is important to get a sense of the error associated with your chosen representation. 


What if you wanted to do a prediction? The average moving average cannot be used since it centers the month in the duration. You would need to use a trailing moving average since it uses historical periods for averaging the values
For purposes of forecasting, we use trailing moving averages, where the window of k periods is placed over the most recent available k values of the series. Set the align = "right" to acquire historical data.
```{r}
savings_tma <- economics %>%
  select(date, srate = psavert) %>%
  mutate(srate_tma = rollmean(srate, k = 12, fill = NA, align = "right"))

```

```{r}
savings_tma %>%
  gather(metric, value, -date) %>%
  ggplot(aes(date, value, color = metric)) +
  geom_line()
```

Moving Average, requires that the weights add to one

```{r}
ma(AirPassengers, order = 12, centre = TRUE)
autoplot(AirPassengers, series = "Data") + 
  autolayer(ma(AirPassengers, order = 12, centre = T), series = "2x12-MA") +
  ggtitle("Monthly Airline Passengers (1949-60)") +
  labs(x = NULL, y = "Passengers")
```