ZZ_Midterm_Mockup.qmd

---
title: "TS_PPLE_2023Spring_Midterm"
toc: true
toc-location: left
toc-depth: 6
self-contained: true
format: html
editor: source
params:
  print_sol: true
---

```{r}
library(fpp3)
```

# 0. Import data

Run the code below. A pop-up window will appear (look for it). Select the file "Spain_Arrivals_Monthly.csv", placed in the ZZ_Datasets folder

```{r}
sp_arrivals <- readr::read_delim(file.choose(), delim = ",") %>% 
            mutate(year = substr(period, 1, 4),
                   month = substr(period, 6, 8),
                   ym = make_yearmonth(year = year, month = month),
                   value = as.numeric(gsub(".", "", value, fixed=TRUE))
                   ) %>% 
            select(ym, value) %>% 
            as_tsibble()
            
sp_arrivals
```

# 1. Basic plots

## 1.1 Create a time-plot of the series, adjusting the time grid so that it signals the beginning of every year

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

## 1.2 Looking at the timeplot prior to 2020, what is the seasonal period you would expect? (max. 30 words)

------------------------------------------------------------------------

YOUR ANSWER GOES HERE (30 words max)

------------------------------------------------------------------------

## 1.3 Looking at the timeplot, judge briefly the strength of the seasonal vs the trend component:

------------------------------------------------------------------------

YOUR ANSWER GOES HERE (30 words max)

------------------------------------------------------------------------

# 2. TS Decomposition

## 2.1 Perform an STL decomposition with default arguments. Depict the decomposition.

Store the resulting components in a variable called `STL_defaults`. Then depict the resulting decomposition.

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

## 2.1.1 If you can, adjust the parameters of the STL decomposition to improve it. Depict the resulting decomposition.

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

## 2.1.2 What are the most important limitations of the STL decomposition in general?

------------------------------------------------------------------------

Bullet point like answer (30-40 words max)

------------------------------------------------------------------------

## 2.1.3 Check, in fact, the decomposition is indeed a breakdown of the time series.

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

## 2.2 Perform a classical decomposition. Store the resulting components in a tsibble called `dcmp_classic`

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```


## 2.3 Compare the STL and Classical decompositions in terms of:

1.  Variance of their components (assess graphically)

------------------------------------------------------------------------

YOUR ANSWER GOES HERE (50 words max)

------------------------------------------------------------------------

2.  Autocorrelation of the remainder / irregular component

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

------------------------------------------------------------------------

YOUR ANSWER GOES HERE (50 words max)

------------------------------------------------------------------------

# 3. Benchmark modes

## 3.1 Filter a subset of the data so that you retain only data up to January 2020. Store it in a new variable called `sp_arrivals_jan2020`

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

## 3.2 Consider two tsibbles: the original `sp_arrivals` and the reduced `sp_arivals_jan2020`. Then fit the following forecasting models to each of these tsibbles seaparately.

1. Seasonal Naive model
2. SES model
3. Drift model
4. `decomposition_model()` using STL for the decomposition (decomposition you used in section 2) + drift() for seasonally adjusted component + seasonal naive for seasoanl component.
5. `decomposition_model()` using STL for the decomposition (decomposition you used in section 2) + drift() for seasonally adjusted component + seasonal naive for seasoanl component.

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

## 3.3 Produce forecasts of up to 1 year ahead with all the models. Store them in two variables called `fc_arrivals` and `fc_arrivals_jan2020`.

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

## 3.4 Depict the forecasts along with the original time series for model 4. of those specified in 3.1 (decomposition_model with drift). Do this for both `fc_arrivals` and `fc_arrivals_jan2020` (two separate graphs).

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

# 4. Assess the residuals of `decomposition_model()` using SES for the seasonally adjusted component that has been fitted to the totality of the time series. For the autocorrelation, be sure to include use of the Ljung-Box or Box-Pierce statistics.

```{r}
#YOUR CODE GOES HERE

# PLACE ALL THE CODE WITHIN THIS SNIPPET
```

------------------------------------------------------------------------

YOUR ANSWER GOES HERE (100 words max)

------------------------------------------------------------------------