Link Search Menu Expand Document
LOST

Creating a Time Series Dataset

Time-series estimators are, by definition, a function of the temporal ordering of the observations in the estimation sample. So a number of programmed time-series econometric routines can only be used if the software is instructed ahead of time that it is working with a time-series dataset.

Keep in Mind

  • Date data can be notoriously difficult to work with. Be sure before declaring your data set as a time series that your date variable has been imported properly.

  • As an example, we will use data on U.S. quarterly real Gross Domestic Product (GDP). To get an Excel spreadsheet holding the GDP data, go to the Saint Louis Federal Reserve Bank FRED website.

Implementations

Python

pandas supports time series data. Here is an example which downloads quarterly data, casts the date column (read in as an object series) as a datetime series, and creates a year-quarter column.

import pandas as pd

# Read in data
gdp = pd.read_csv("https://github.com/LOST-STATS/lost-stats.github.io/raw/source/Time_Series/Data/GDPC1.csv")

# Convert date column to be of data type datetime64
gdp['DATE'] = pd.to_datetime(gdp['DATE'])

# Create a column with quarter-year combinations
gdp['yr-qtr'] = gdp['DATE'].apply(lambda x: str(x.year) + '-' + str(x.quarter))

R

There are many different kinds of time series data set objects in R. Instead of R-based time series objects such as ts, zoo and xts, here we will use tsibble, will preserves time indices as the essential data column and makes heterogeneous data structures possible.

The tsibble package extends the tidyverse to temporal data and built on top of the tibble, and so is a data- and model-oriented object.

For more detail information for using tsibble such as key and index, check the tsibble page and the Introduction to tsibble.

STEP 1) Load necessary packages

# If necessary
# install.packages(c('tsibble','tidyverse'))
library(tsibble)
library(tidyverse)

STEP 2) Import data into R.

gdp <- read.csv("https://github.com/LOST-STATS/lost-stats.github.io/raw/source/Time_Series/Data/GDPC1.csv")

# read.csv() has read in our date variable as a factor. We need a date!
gdp$DATE <- as.Date(gdp$DATE)
# If it were a little less well-behaved than this, we could use the lubridate package to fix it.

STEP 3) Convert a date variable formats to quarter

gdp_ts <- as_tsibble(gdp,
                     index = DATE,
                     regular = FALSE) %>%
    index_by(qtr = ~ yearquarter(.))

By applying yearmonth() to the index variable (referred to as .), it creates new variable named qtr with a quarter interval which corresponds to the year-quarter for the original variable DATE.

Since the tsibble handles regularly-spaced temporal data whereas our data (GDPC1) has an irregular time interval (since it’s not the exact same number of days between quarters every time), we set the option regular = FALSE.

Now, we have a quarterly time-series dataset with the new variable date.

References for more information:

  1. If you want to learn how to build various types of time-series forecasting models, Forecasting: Principles and Practice provides very useful information to deal with time-series data in R.
  2. If you need more detail information on tssible, visit the tsibble page or tsibble on RDRR.io.
  3. The fable packages provides a collection of commonly used univariate and multivariate time-series forecasting models. For more information, visit fable.

Stata

STEP 1) Import Data to Stata

import delimited "https://github.com/LOST-STATS/lost-stats.github.io/raw/source/Time_Series/Data/GDPC1.csv", clear

STEP 2) Generate the new date variable

generate date_index = tq(1947q1) + _n-1

The function tq() converts a date variable for each of the above formats to an integer value (starting point of our data is 1947q1).

_n is a Stata command gives the index number of the current row.

STEP 3) Index the new variable format as quarter

format date_index %tq

This command will format date_index as a vector of quarterly dates which corresponds to our original date variable observation date.

STEP 4) Convert a variable into time-series data

tsset date_index

Now, we have a quarterly Stata time-series dataset. Any data you add to this file in the future will be interpreted as time-series data.